KR102199412B1 - Data Mining System Using dimensionless number and the method of the same - Google Patents

Abstract

It relates to a data mining system (1) using dimensionless numbers. Such a data mining system 1 comprises: a unit converter 3 for converting a physical quantity into a dimensionless number to set a parameter (Mother number); An embroidery generator 5 for setting analysis data by generating a child number by converting a parameter set by the unit converter 3 according to a certain rule; A database (13,15) for storing and fetching the generated parameter and embroidery, and a relational expression matching the parameter and embroidery; It includes a mining unit 7 for searching, transforming, modeling, evaluating, and predicting by interworking with the databases 13 and 15 to retrieve analysis data.

Description

Data Mining System Using dimensionless number and the method of the same}

The present invention relates to a data mining system using non-dimensional numbers, and more specifically, by converting relational expressions such as experimental data and equations into non-dimensional numbers, and analyzing the data composed of the non-dimensional numbers by a mining technique, the accuracy of the results. And a data mining system capable of increasing reliability and discovering links between different technical fields.

In general, data mining is defined as extracting useful information from a large data set. In more detail, data mining is used to explore and analyze large amounts of data using automated or semi-automated tools to discover meaningful patterns and rules. It's a process.

As such data mining is used in various fields, in the security field, it is used to determine whether each packet has a threat to the network. In the medical field, data mining is used to predict the recurrence of cancer. In the field, data mining is used to find out what factors affect the accuracy of the missile, and in the economic field, it is used to predict the economic outlook.

Data mining is typically performed in the following order.

First, we acquire data to be used for the purpose of data mining project identification and analysis.

This step involves randomly sampling a large number of databases to identify the records to be used for analysis. Here, the database may be internal data, such as customer's past purchase data, or external data, such as credit rating data. In general, data mining deals with very large databases, usually requiring thousands or tens of thousands of records.

When data extraction is complete, the data is explored, refined, and preprocessed.

This step is to verify that the data is in reasonable conditions as follows.

In other words, it is verified whether the data value is within a reasonable range and whether the extreme value exists when the method of handling the missing value among the data and the expected value for each variable is given.

For example, use a graph such as a matrix table of a scatterplot showing the relationship between each variable and all other variables. Also, check the consistency of the definition of the variable, the measurement unit, and the measurement period.

And, if necessary, the data is reduced, and in the case of supervised learning, the data is divided into a data set for learning, evaluation, and verification.

In this step, unnecessary variables are removed and variables are converted. For example, convert expenses into expenses in excess of $100 and less than $100. Then, create a new variable.

For example, create a variable that tells you whether you have purchased at least one of several products. Also, check what each variable means and whether it makes sense to include the variable in the model.

After these steps are completed, data mining tasks are classified, predicted, and clustered.

Classification is the most basic form of data analysis. For example, in sales data, a person who has received a sale or purchase offer is classified as a person who responds or does not respond to it, or a loan applicant is classified as a person who pays off money on time or late or declares bankruptcy.

In addition, it is also possible to identify rules by using similar data with known classification results, and then apply the rules to data for which the classification results are unknown.

Prediction predicts the value of a numeric variable, such as a quantity of purchase, rather than a variable of a group, such as buyers or non-buyers.

In this step, you select an analysis type that suits the purpose of the data mining project.

Then, select the data mining techniques to be used, such as regression analysis, neural network model, and hierarchical cluster analysis.

In other words, when the purpose of data mining is prediction, multiple linear regression analysis, regression tree, neural network model, and K-nearest neighbor method are selected. In case of classification, logistic regression analysis, classification tree, neural network model, discriminant analysis, K -Select the nearest neighbor method and the simple Bayesian classification model method.

When a data mining technique is selected in this way, data mining is performed by applying an algorithm.

This step is generally an iterative process, and various variables are applied, such as differently applying explanatory variables or detailed selection conditions of the algorithm within one algorithm. And. When these conditions are appropriate, use variables that are appropriately improved by receiving feedback from the performance of the algorithm using the evaluation data.

The process of data mining as described above will be described as an example when purchasing a product in the market.

The process of purchasing a product in the market and putting it in the shopping cart corresponds to the selection of data, and excluding a product that has purchased only one of the items in the shopping cart because it cannot compare similarities with other items corresponds to refining the data. Transforming values into specific code values or sequence values, for example, designating milk and cheese as A0001, MK, CS corresponds to data transformation, and is used in data mining that analyzes the relationship between transformed milk and cheese. It corresponds.

In addition, this data mining technique is applied to various fields, for example, it can be applied to physics, chemistry, and pharmaceutical fields.

In other words, the most representative method of these is data mining using artificial intelligence technology with an open genome database. In other words, by using the published gene expression (transcriptome) database, comparing with normal cells and tissues, analyzing how specific gene expression patterns change in cells and tissues related to specific diseases, and expressing in specific diseased cells and tissues. It is a strategy to increase the success rate of new drug development by screening in a large-scale compound library for compounds necessary to convert the pattern into a pattern of normal cells and tissues.

However, the conventional data mining system has a problem in that the accuracy of the experimental data and constant values used as basic data is low, and the accuracy of the embroideries calculated therefrom is also low, so that the result is not reliable.

Patent Application No. 10-2000-38050 (Name: Closest Query Processing Method Using Spherical Pyramid Technique) Patent application No. 10-2015-7019601 (Name: method for treating disease) Patent application No. 10-2017-39140 (Name: stroke prediction and analysis system and method) Patent application No. 10-2003-25426 (Name: test execution system and its driving method)

Accordingly, the present invention has been devised to solve such a problem, and an object of the present invention is to convert relational expressions such as experimental data and equations into dimensionless numbers, and analyze the data composed of the dimensionless numbers by a mining technique. It is to provide a data mining system and method that can increase the accuracy and reliability of the data and discover links between different technical fields.

In order to achieve the above object, an embodiment of the present invention,

A unit converter 3 for converting a physical quantity into a dimensionless number to set a mother number;

An embroidery generator 5 for setting analysis data by generating a child number by converting a parameter set by the unit converter 3 according to a certain rule;

A database (13,15) for storing and fetching the generated parameter and embroidery, and a relational expression matching the parameter and embroidery;

A data mining system 1 using a dimensionless number is provided, including a mining unit 7 that performs search, transformation, modeling, evaluation, and prediction by interworking with the databases 13 and 15 to retrieve analysis data.

Another embodiment of the present invention relates to a method for data mining a dimensionless number using the above-described data mining system 1,

A first step (S100) of converting a physical quantity into a dimensionless number by the unit converter 3 to set a parameter (Mother number);

A second step (S110) of generating an embroidery (child number) by converting the parameter set by the unit converter 3 according to a predetermined rule by the embroidery generator 5 to set analysis data;

A third step (S120) of storing and retrieving the generated parameter and embroidery, and a relational expression for matching the parameter and embroidery in the databases 13 and 15; And

A fourth step (S130) in which the mining unit 7 interworks with the databases 13 and 15 to retrieve the analysis data to perform search, transformation, modeling, evaluation, and prediction.

Another embodiment of the present invention includes a recording medium in which the data mining method using the dimensionless number is recorded as a program and readable by an electronic device.

The data mining system according to an embodiment of the present invention converts relational expressions such as experimental data and equations into dimensionless numbers, and analyzes the data composed of the dimensionless numbers by a mining technique, thereby increasing the accuracy and reliability of the results. In addition, it has the advantage of discovering links between different technical fields.

1 is a diagram schematically showing an operation process of a data mining system using dimensionless numbers according to an embodiment of the present invention.
FIG. 2 is a block diagram schematically showing the structure of a data mining system using a dimensionless number shown in FIG. 1.
3 is a diagram schematically showing the structure of a unit converter of the data mining system shown in FIG. 2.
4 is a flow chart showing a process of analyzing an input physical quantity by converting it into a dimensionless number.
5 is a block diagram schematically showing the structure of the mining unit shown in FIG. 2.
6 is a block diagram schematically showing the structure of the modeling unit shown in FIG. 5.
7 is a flowchart sequentially showing a data mining method using dimensionless numbers according to another embodiment of the present invention.
FIG. 8 is a flow chart specifically showing a data mining step, which is a fourth step of the data mining method illustrated in FIG. 7.

Hereinafter, a data mining system using dimensionless numbers according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

The data mining system 1 proposed by the present invention converts physical quantities to non-dimensional numbers, and analyzes the results of non-dimensional data by a data mining technique consisting of sampling, searching, transformation, modeling, evaluation, and prediction. It relates to a technology that can improve accuracy and reliability, and discover links between different technical fields, and relates to a so-called Buckingham machine.

In addition, the dimensionless number applied in the present invention is a concept obtained by converting all of the formulas, equations, and empirical formulas of physics into numerical values. Through this, the SI units are compatible with each other, and the authenticity of the equation including this unit can be easily verified. I can.

Explaining these dimensionless numbers, the scientific community is based on the mass (kg), length (m), time (s), luminous intensity (cd), mass (mol), current (A), temperature (K) There are 7 international units in the world, and these units are artificially made and thus are improved according to the times.

Specifically, the present invention establishes the first post of so-called zero zone theory in order to unify these seven units and to be compatible with each other.

This post refers to units decided by the International General Assembly of Weights and Measures (CGPM), and a new SI is created by fixing the values of seven defining constants.

As it is a positive constant, the uncertainty is 0 and has exact values. The symbols of the seven constants are frequency Δv _Cs , light velocity c, Planck's constant h , electron charge e , Boltzmann constant k _B , Avogador constant N _A , Using luminous efficacy K _cd , etc., the first postulate of the zero zone theory is as follows.

Δv _Cs = c = h = e = k _B = N _A = K _cd = 1

In particular, the number 1 means 1 photon, and it is assumed that 1 photon and the speed of light, Planck's constant known as the smallest unit of energy in the universe, and 1 second are equivalent to each other.

At this time, the number of photons is a number, it becomes the natural frequency of the phenomenon at every moment, and since the frequency has the same meaning as energy, this frequency can be quantified in nature with continuity to the number 1.

For example, it can be expressed as E (energy) = h (Planck's constant) * v (frequency), assuming that h = 1, the above equation holds E = v , which means that the energy and the frequency are the same. do.

And, when each unit is converted to a dimensionless number using the zero zone first postulate, it is shown in Table 1 below.

Item Basic unit symbol Zero zone dimensionless number time 1s 9192631770 Length 1m 30.663318988... mass 1Kg 1.475521399...*10 ⁴⁰ electric current 1A 6.789686817...*10 ⁸ Thermodynamic temperature 1K 2.266665264... Substance 1mol 6.02214076*10 ²³ brightness 1cd 2.614830482...*10 ¹⁰

In Table 1, the process of converting meters (m), which is a unit of length, to a dimensionless number is described, as the light flux c=299792458m/s, s=9192631770, and assuming c=1 by the postulate, the length ( The dimensionless number of m) can be expressed as 30.663318988...

In addition, when explaining the process of converting Kg, which is a unit of mass, to a dimensionless number, Planck's constant h=6.62606896*10 ^-34 Js, and when Joule (J), which is an energy unit, is expressed as a basic unit, h=6.62606896*10 ^-34 Kg *m ² /s. At this time, since it is defined as s=9192631770 and h=1 by the postulate, the dimensionless number for the length (m) is calculated as above, and thus the dimensionless number of the mass (Kg) is 1.475521399...*10 ⁴⁰ .

In this way, other units such as current, thermodynamic temperature, mass, and brightness can also be converted into dimensionless numbers.

In addition, since the above 7 units are converted into dimensionless numbers, all derived units derived from these units can also be converted into dimensionless numbers. For example, derived units such as Hertz (Hz), amount of charge (C), Newton (N), and Pascal (Pa) can be converted into dimensionless numbers through a combination of defined basic units.

In addition, it is possible to calculate units of different dimensions by such a dimensionless number. For example, in the case of calculating 1kg and 1m, or calculating 1 second and 1℃, it is possible by converting each unit into a dimensionless number and then arranging the dimensionless number with each other.

In addition, since each basic unit or derived unit can be expressed as a number, the number can be expressed as an equation in reverse, so that the authenticity of the equation can be verified by a simple operation rather than an experiment.

For example, in the case of the formula F=ma, there are many ways to verify that this expression is correct or incorrect, but the key is to satisfy the equal sign (=) within a given uncertainty in the measurement.

If the dimensionless number of the present invention is applied to the force (N), mass (kg) or acceleration (a) described in units on the left and right sides and converted into a numerical value, only the number will remain on the left side or the right side, within uncertainty whether the two are the same number. This is a method of verifying the authenticity of the above equation by simply checking it.

In the present invention, by analyzing such a dimensionless number by data mining, it is possible to grasp a correlation relationship or a causal relationship for data in a more diverse field.

In more detail, as shown in Figs. 1 to 6, the data mining system 1 using a dimensionless number according to the present invention,

A unit converter 3 for converting the input physical quantity into a dimensionless number to set a parameter (Mother number);

An embroidery generator 5 for setting analysis data by generating a child number by converting a parameter set by the unit converter 3 according to a certain rule;

A database (13,15) for storing and fetching the generated parameter and embroidery, and a relational expression matching the parameter and embroidery; And

And a mining unit 7 that fetches analysis data in connection with the databases 13 and 15 to perform search, transformation, modeling, evaluation, and prediction, and output through the output unit 12.

In this data mining system (1),

The unit converter (U2CA) 3 converts measured values, empirical formulas, relational expressions, equations, experimental data, etc. (hereinafter referred to as relational expressions) in each field such as mathematics, physics, and chemistry into dimensionless numbers to set parameters.

The unit converter 3 having such a configuration converts the relational expression to a dimensionless number by applying the first postulate of the zero zone theory as described above to construct initial data (Initiative Data).

Accordingly, the unit converter 3 includes an input unit 9 receiving a physical quantity in order to construct initial data in a dimensionless number; An operation unit 10 that makes the physical quantity dimensionless by replacing the unit of the input physical quantity with a dimensionless number by a so-called zero zone code (zero zone first postulate); It includes an output unit 12 for outputting the dimensionless physical quantity.

In the unit converter 3 having such a configuration, measured values, empirical formulas, equations, etc. in each field such as mathematics, physics, and chemistry are input to the input unit 9.

Then, the input relational expression is converted into a dimensionless number by the operation unit 10.

In addition, the converted dimensionless number is set as a parameter and is expressed in the form shown in Table 2 below.

That is, a parameter is generated by converting the relational expression described in the image column of Table 2 into a numerical value through a compatible code.

Then, these parameters are converted into embroidery by applying a certain rule through the embroidery generator 5.

The embroidery generator 5 performs a function of searching for an optimal proximity value when a user inputs a query.

In more detail, the embroidery generator 5 generates embroidery expressed in the form shown in Table 3 below.

As described above, embroidery is calculated through compatible codes and generation rules.

At this time, the parameter generation rules are shown in Table 4 below.

Operator Explanation * 2~99 Multiply by 2~100 (gen_cn_basic) * 1 / 13~100 Sharing from 13 to 100 (gen_cn_basic) * 1 ~ 12/1 ~ 12 1~12 / Multiply 1~12 (gen_cn_basic) * 10^1~100 Multiply by 10^1~100 (gen_cn_basic) ^ (2~100/2~100) 2~100 / 2~100 squared (generateChildNumSmall) ^ (x ^ a) / (y ^ b) Find 100 values between 1.001 and 1.002001 (gen_cn_basic_power) * Math Multiply basic factors (ln, log, pi, pi^2, e, e^2, etc.)

Therefore, the embroidery shown in Table 3 can be calculated by applying the compatible code linked to the parameters in Table 2 and the operator in Table 4.

For example, 9.7851306871120166801272960757596, which is the embroidery of item 1 in Table 3, can be created by applying the operator In to the parameter 1.0000009785135474552706397491205 of item 3 in Table 2.

In the end, the embroidery of item 1 is composed of the parameter *In, and In serves as the derivation path information.

Therefore, it is possible to trace the parameter from the embroidery by inverting the embroidery of this type in the analysis step of data mining to be described later.

And, the generated embroidery is stored in a conversion DB (Convergency DB). In the conversion DB, parameters and embroidery are stored respectively. In addition, equations, constants, empirical equations, and mining equations that match parameters and embroidery are stored in a reference DB.

In addition, the above-described parameters and embroidery are interlocked with each other through a compatible code, and such correlations are configured in Table 5 below, for example.

Referring to Table 5, a field containing a dimensionless number of a parent relational expression (mother_number), a field containing a reference code of the parent relational expression (equation_address), a field containing a mathematical operation method applied to obtain a child relational expression from the parent relation mathematical_operation), and a field (child_number) that contains a non-dimensional number (embroidery) obtained through a mathematical operation method for a non-dimensional number (parameter) of the parent relation.

Meanwhile, a physical quantity can be analyzed using this compatible code table.

That is, the present invention includes a search module 17, and after the search module 17 receives a physical quantity to be analyzed, it interworks with the databases 13 and 15 to replace the unit of the physical quantity with a zero zone code, It converts to, searches for a compatible code corresponding to the non-dimensionalized physical quantity, and searches the compatible code to identify the record containing the corresponding non-dimensional number.

In more detail, as shown in FIG. 4, when a user inputs a physical quantity to be analyzed through an input window popped up on the terminal screen (S10) and requests analysis, the search module 17 It checks whether it is included (S20).

If there is a unit, the unit is replaced with a zero zone code and converted into a dimensionless number (S30). Conversely, if there is no unit, the process immediately proceeds to step S40.

Then, in step S40, a compatible code corresponding to the dimensionless physical quantity is searched. At this time, the field to be searched is embroidery.

After the search, it is determined whether there is a dimensionless number without errors, or if there is an error, 10 analysis results are output in ascending order and descending order, respectively, based on the embroidery in the order of a certain error size (S50). This process corresponds to the ranking algorithm.

That is, if there is a non-dimensional number without error, the compatible code is searched to identify the record containing the non-dimensional number, and then the parameter (mother_number), the reference code of the relational expression (equation_address), and the mathematical operation method (mathematical_operation) are determined from the record. After reading it, print the result.

In this case, if the result is as follows, it can be interpreted as follows.

(Result) 4.2168740563618544990834786089658e+42

=(P-197-4-1-10)^97*55=(2.64069781000404991)^97*55

The input value 4.2168740563618544990834786089658e+42 is the dimensionless physical quantity to be analyzed. And, P-107-4-1-10 is a reference code of a relational expression related to this physical quantity.

^97*55 is a mathematical calculation method for deriving 4.2168740563618544990834786089658e+42 from the relation referred to as P-107-4-1-10.

Further, 2.64069781000404991 is a dimensionless number corresponding to the relational expression referenced from P-107-4-1-10.

Therefore, it can be seen from the above operation result by the search module 17 that the number of objects to be searched is derived by multiplying the relational expression referred to as P-107-4-1-10 to the power of 97 and multiplying by 55, and The nondimensionalized physical quantity 4.2168740563618544990834786089658e+42 can be confirmed by the relational expression referred to as P-107-4-1-10 and the mathematical operation method.

Meanwhile, the generated embroidery can be analyzed by the mining unit 7. The mining unit 7 inputs a query to the generated embroidery and extracts the output result by a ranking weight method to obtain target data, and analyzes the correlation and causal relationship between the target data. Through this, you can grasp the links between each field.

As shown in FIG. 5, the mining unit 7 includes a data search unit 20 for searching for nearest data among dimensionless data stored in a DB by an input query; A conversion unit 22 for converting data to be suitable for analysis; It includes a modeling unit 24 that selects some of the converted data as target data, and performs correlation analysis, cluster analysis, classification analysis, and numerical division analysis.

The process of mining data by the mining unit 7 will be described in more detail in the order of data search, data transformation, modeling, evaluation and prediction.

The data search unit 20 performs data search, and the data search identifies dimensions, sizes, scales, attributes, etc. of the generated embroidery data.

That is, the test data composed of the generated embroidery is stored in the conversion DB 13, and the user enters and searches a query including a certain condition to obtain a target result. At this time, the input query is SQL( structured query language), GIS, etc., through a query function, the closest data is searched among the test data stored in the DB 13 during conversion.

For example, assuming that the test data table is shown in Table 6 below, the nearest neighbor data is searched using the K-nearest neighbor search method.

mass energy ... Temperature Unprecedented Component A 6.3077878*E24 6.3123878*E24 6.6757878*E24 6.3345878*E24 Component B 6.1237878*E24 6.3324878*E24 ... 6.3345878*E24 6.3078978*E24 Ingredient C 6.3457878*E24 6.4567878*E24 ... 6.3768878*E24 6.3077890*E24
.
.
. .
.
.
... .
.
.
.
.
.
Ingredient D 6.4227878*E24 6.7657878*E24 ... 6.3078978*E24 6.3077678*E24 Ingredient E 6.4537878*E24 6.7867878*E24 ... 6.3123878*E24 6.3075678*E24

In the test data, when searching for a component closest to the condition by searching for each component in the order of mass and temperature, the mass and temperature are entered in a query input window, and each component is arranged in order by a function.

First, a data set is loaded in order to find a component most similar to a reference component based on mass and temperature in the data set. For example, data is loaded with the following knn search function syntax.

load chemical data set;

X = [component.mess component.temp.];

Y = [20 162; 30 169; 40 168; 50 170; 60 171];

In addition, the nearest data is searched among the arranged mass and temperature data. For example, the search is performed through the following function syntax.

Idx = knnsearch(X,Y)

Idx = knnsearch(X,Y,Name,Value)

[Idx,D] = knnsearch(___)

In the above function syntax,

Idx = knnsearch(X,Y) finds the nearest neighbor in X for each query point in Y, and then returns the index of the nearest neighbor as an Idx column vector, and the number of rows of Idx is equal to Y.

And, Idx = knnsearch(X,Y,Name,Value) returns Idx by additionally specifying one or more name-value pair arguments as an option. For example, the number of nearest neighbors to be searched and the number used for search You can specify the distance measurement method.

Also, [Idx,D] = knnsearch(___) additionally returns a matrix D using the input arguments of the syntax listed above, where D is the distance between each observation in Y and the closest observation in X Includes.

And, the result of searching for the component closest to the reference component based on the mass and temperature in X is output as follows.

X(Idx,:)

ans = 2ㅧ1

Component A 6.3077878*E24 6.6757878*E24

Component C 6.3457878*E24 6.3768878*E24

In addition, since the searched nearest data is analyzed, various analysis methods are possible, and analysis is performed by, for example, a ranking weight method.

At this time, in order to perform the analysis on the nearest data, the data is appropriately converted so as to be suitable for analysis, and redundant data is removed by the conversion unit 22, and normalization and dimension reduction are performed.

Dimension reduction is a technique for reducing the amount of data when analyzing data, and it is difficult to analyze data when there are many features of the data, especially when there are three or more (three-dimensional) features. In the case of machine learning, when there are many features of the learning data, the computational amount increases, and in particular, more data is required for learning. In order to reduce such a large amount of data, dimension reduction is performed.

There are two types of dimensionality reduction techniques: feature selection and feature extraction. In the case of feature selection, it is a method of selecting only a few of the main fields that best represent the characteristics of the data among several features and selecting them as representative features.

For example, if you have [7,1,2],[100,1,3],[92,1,5], since the first and third columns of these three matrices have the largest change width, the first and Using only the third column as a representative feature, select as follows. [7,2],[100,3],[92,5]

Next, there is a feature extraction method, which extracts data in a completely different form from the original data. For example, it is a method of expressing characteristics by converting [7,1,2] into [3,4] etc. by a certain formula.

The dimensionality reduction can be performed by various algorithms. For example, the dimensionality reduction is possible with the following algorithm.

In this way, after data is converted, modeling is performed by the modeling unit 24. In other words, modeling is a process of selecting some of the converted data as target data, and analyzing the association between this target data and other data, cluster analysis, classification analysis, and numerical division analysis.

As shown in FIG. 6, the modeling unit 24 includes a numerical division analysis module 30; A correlation analysis module 32; Causal analysis module; Cluster analysis module 36; And evaluation prediction module 38.

First, the process of analyzing data by the numerical division analysis module 30 will be described as follows.

For example, if the data to be analyzed is 0.123456789012345678001234567890023456734, numerical division is performed through the following process.

First, analysis rules 1 to 5 stored in an analysis rule table are loaded to sequentially divide the analysis target data according to a predetermined rule.

That is, 123456789 is sequentially divided at the decimal point as the first step, and then 12345678 is divided as the second step with the boundary of 0, and then 12345678 is divided as the third step with the boundary of 00, and then 00 As the fourth step, 23456734 is divided. As a result, the data to be analyzed can be divided as follows.

By dividing the data to be analyzed multiple times as described above, the limit of the search data is set, and the following data is ignored.

In addition, the numerically divided data is stored in the embroidery (CN) and reference formula (Ref.) and merge analysis history DB stored in the convergence DB, and the format is as follows.

That is, the first step = embroidery 1 + Ref.1, the second step = embroidery 3 + Ref.5, the third step = embroidery 1 + Ref.2.

On the other hand, it is possible to analyze the relationship between target data and other data as well as numerical division.

That is, the association analysis module 32 analyzes the association between data, and the association is a kind of clustering that groups information with a lot of association by grasping the association of information through data mining, creating valuable information and applying it to decision making. This is the analysis method you want to do This association is different from predictions for obtaining an answer (for example, yes/no) that has not yet occurred to a specific problem or segmentation of objects according to a specific purpose.

Taking this association analysis as an example, it is to analyze the association between components A and B, and analyzes whether component B reacts when component A reacts.

That is, when component A reacts, when component B reacts, it is judged to be highly related. [At this time, A is lhs (left-hands side), B is rhs (right-hands side)]

And, the indicators of the association analysis include support, confidence, and lift.

Support refers to P(A∩B), which is the probability that component A and component B will react simultaneously. The higher the support, the higher the relevance.

Reliability refers to the probability that component B will react when component A reacts, that is, P(B|A). However, the criterion is sometimes unclear because the result depends on component A.

In this case, the degree of improvement is confirmed,

The degree of improvement is the ratio of the probability that component B will react when component A reacts and the probability that component B will react regardless of the reaction of component A.

In this degree of improvement, if component A and component B are independent events, then P(B∩A) is P(A)*P(B), so LITF = P(A)*P(B) / P(A)*P( B) = 1.

That is, if A and B are not related, LIFT=1, and if LIFT> 1, the greater the LIFT value, the higher the correlation. If LIFT <1, it means that if component A reacts, component B does not react.

In the case of actually performing this association analysis, it proceeds with the following algorithm.

First, as in items 1 and 2, the data file to be analyzed is loaded.

And, in order to divide into reactive components like item 4, that is, in order to convert into a transaction structure, the components must be organized and the format must be divided into a list, so a split function (Split function) is used.

And, like item 6, it is converted into a transaction structure.

And, like item 10, it is converted into data.frame format, and as a result, the following data is displayed.

As shown in the output data, it can be seen that component D has the highest reaction frequency, and component B and component C react together.

Therefore, it can be seen that component D has a high correlation with components B and C.

In addition, the data for which the correlation is analyzed may be fed back to the unit converter 3 and set as a parameter again.

On the other hand, in the modeling of the present invention, it is possible to analyze causality as well as correlation between target data.

That is, the causal relationship is analyzed by the causality analysis module 34, and the causal relationship is a linear prediction of the variable X by combining the past data of one variable X and the past data of the other variable Y in two time series data. When linear regression) was performed, if the linear prediction with only the past data of the other variable Y was statistically significant and helped in the prediction, X can be defined as the cause of Y, and this relationship is a causal relationship.

In order to check the causal relationship, two time series variables, that is, two sets of time series data, are required, and a time difference (lag or delay) must be entered as a parameter.

In other words, the parallax is used to determine whether A affects to the point in time after B when testing two time series data sets A and B, or vice versa. It's a process.

At this time, the set of past data of A is the parallax of A, and the set of past data of B is the parallax of B.

This causal analysis is a prerequisite, and both data to be tested must satisfy stationary.

Further, as an additional prerequisite, the direction of the test must be considered. When there are two variables A and B to be tested, a total of two tests must be performed in both directions as a set.

That is, with the direction of variable A → variable B, the lags determined by the test are input and the causal relationship test is performed.

Through two tests, we test whether A has a causal effect on B and whether B has a causal effect on A. The combination of these results results in four cases.

That is, if A gives a causal effect on B, B does not have a causal effect on A, B gives a causal effect on A, and A does not have a causal effect on B, A gives a causal effect on B. , B also has a causal effect on A, A does not have a causal effect on B, and B also does not have a causal effect on A.

When explaining such a causal relationship analysis as an example, whether or not the variable A affects B is examined by the following algorithm.

First, run the R program, and analyze the causal relationship between whether B affects A by the granger test function. First, enter 4 as the Lag value as shown in the function formula below.

Executing this function gives the following result.

From the output result, it can be seen that the p-value is 0.006414, so that the causal relationship is established because B is the causal factor of A.

Conversely, looking at whether A affects B, the calculation is reversed from the above equation.

As can be seen from the output result, since the p-value is 0.8881, it can be seen that the causal relationship is not established because A is not the causal factor of B.

In addition, the data for which the causal relationship is analyzed as described above may be fed back to the unit converter 3 and set as a parameter again.

Meanwhile, for modeling, it is also possible to cluster and analyze target data.

Cluster analysis, as analyzed by the cluster analysis module 36, means classifying data into clusters having similar characteristics.

This cluster analysis is mainly used in the exploratory analysis stage in the initial stage of analysis, and consists of only input data without output data, and is mainly used to understand or understand the characteristics of data.

This cluster analysis is carried out by two principles, maximizing the degree of aggregation within the cluster and maximizing the degree of separation between clusters.

Maximizing the degree of aggregation within a cluster is to maximize the distance between the same clusters, so that the distance within the cluster is minimized, and maximizing the degree of separation between clusters maximizes the distance between different clusters.

That is, within the same cluster, the distance between variables is close, and the distance between different clusters is far away, which means that clustering is high.

Types of distance scales for this clustering include Euclidean distance, Manhattan distance, standardized distance, and Minkowski distance.

In addition, the types of clustering analysis are divided into separate (non-hierarchical) clustering and hierarchical clustering. An example of a separate clustering is K-Means Clustering, and an example of hierarchical clustering is H-Clustering.

For example, an algorithm for cluster analysis by K-Means Clustering will be described as follows.

That is, as shown in No. 1 of the above drawing, the k value is input as an initial value, the data is divided into k initial clusters, and the center points of k initial clusters are set as in No. 2, and the red and green circles are the center points. do.

As in step 3, the distance between the points of each data entity and the current cluster center point is calculated.

As shown in the result of operation 4, points are clustered with the center point closer to each other. If an individual is close to the current cluster average, it is included in the current cluster.

If not, it is included in another cluster, such as #5.

And, as in No. 6, when the average distance of individual clusters is calculated again and the center point of the cluster is recalculated, clustering is performed based on the center point.

And, as in Nos. 7 and 8, even if the clustering process is repeated, the algorithm is completed when there is no point where the cluster is no longer reassigned.

In addition, the cluster-analyzed data may be fed back to the unit converter 3 and set as a parameter again.

At this time, the unit converter 3 includes the determination unit 11 to determine whether the data fed back by the mining unit 7 is a dimensionless number.

That is, as described above, since data input to the input unit 9 of the unit converter 3 is a general relational expression, equation, or equation, data in a format including units or processed by the mining unit 7 and fed back Since data is a dimensionless number format, the unit converter 3 classifies and processes it.

Accordingly, when the determination result is a non-dimensional number, it is directly inputted to the operation unit 10, and when it is not a non-dimensional number, it is inputted through the input unit 9.

In this way, by separately calculating the data fed back by the mining unit 7 from non-dimensional data, it is possible to efficiently process large amounts of data.

Meanwhile, evaluation and prediction is a process of predicting a future event based on the data processed through the above process.

As such evaluation and prediction are analyzed by the evaluation prediction module 38, various methods can be applied to the prediction method, and supervosed prediction (directed knowledge discovery) and unsupervised prediction (Unsupervised prediction) can be applied depending on the presence of a target variable. undirected knowledge discovery).

In map prediction, the data for analysis is composed of n cases and is associated with each case and predicts the target value using input variables (predictive variable, explanatory variable, independent variable) and target variables (response value, result value, dependent variable). Calculate the model

Such map prediction techniques include discriminant analysis, regression analysis, decision tree analysis, neural network analysis, and time series analysis.

In addition, autonomous prediction refers to an analysis method in which the target variable is not clearly defined, and aims to find various types of features that exist in the data rather than finding rules that determine causal relationships. There is cluster analysis in autonomous prediction.

Alternatively, the accuracy can be evaluated by dividing the data into training data, verification data, and evaluation data, and comparing the predicted value from each data using a model obtained using the training data.

Hereinafter, a data mining method using a dimensionless number according to a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

As shown in FIG. 7, in the data mining method using a non-dimensional number proposed by the present invention, a first step of setting a parameter (Mother number) by converting a physical quantity into a non-dimensional number by a unit converter 3 (S100) Wow;

A second step (S110) of generating an embroidery (child number) by converting the parameter set by the unit converter 3 according to a predetermined rule by the embroidery generator 5 to set analysis data;

A third step (S120) of storing and retrieving the generated parameter and embroidery, and a relational expression for matching the parameter and embroidery in the databases 13 and 15; And

A fourth step (S130) in which the mining unit 7 interworks with the databases 13 and 15 to retrieve the analysis data to perform search, transformation, modeling, evaluation, and prediction.

In the data mining method using such a dimensionless number,

In the first step (S100), the input unit 9 of the unit converter 3 receives the physical quantity, and the unit of the physical quantity inputted by the operation unit 10 is replaced with a dimensionless number by a zero zone code to make the physical quantity dimensionless. , The output unit 12 outputs the dimensionless physical quantity.

In other words, by calculating measured values, empirical formulas, relational formulas, equations, and experimental data in each field such as mathematics, physics, chemistry, etc., through the unit converter 3, the first postulate of the zero zone theory is applied and converted into a dimensionless number and initial data Build (Initiative Data).

Then, in the second step (S110), the constructed parameter is converted by the embroidery generator 5 to generate an embroidery (Child number).

At this time, embroidery is created by applying certain rules to the parameters.

For example, by applying the operator In to the parameter 1.0000009785135474552706397491205, the embroidery 9.7851306871120166801272960757596 can be generated.

At this time, since embroidery is composed of a parameter + In, In plays the role of the derived path information.

Therefore, it is possible to trace the parameter from the embroidery by inverting the embroidery of this type in the analysis step of data mining to be described later.

When this second step (S110) is completed, the third step (S120) proceeds.

In the third step (S120), the generated parameter and embroidery, and a relational expression matching the parameter and embroidery are stored and retrieved in the databases 13 and 15.

That is, the generated embroidery and parameters are stored in a conversion DB (Convergency DB) 13. In addition, equations, constants, empirical equations, and mining equations that match parameters and embroidery are stored in a reference DB (15).

In this case, the parameter and the embroidery are stored in a certain format, and include a field (mother_number) in which a dimensionless number of the parent relationship is recorded; A field in which a reference code of a parent relational expression is stored; A field containing a mathematical operation method applied to obtain a child relational expression from the parent relational expression; It has the form of a field (child_number) that contains the number of numbers obtained through a mathematical operation method for a parameter.

Then, the fourth step (S130) proceeds. In this step, the mining unit 7 fetches the analysis data in connection with the databases 13 and 15 to perform search, transformation, modeling, evaluation, and prediction.

That is, the fourth step (S130), as shown in Figure 8,

A step (S60) of searching the nearest data among test data stored in the DB by the data search unit 20 of the mining unit 7 according to the input query; Converting the searched data by the conversion unit 22 (S65); Selecting some of the converted data as target data by the modeling unit 24, performing a correlation analysis, cluster analysis, classification analysis, and numerical division analysis (S70), and evaluation and prediction It includes a step (S80).

Further, a fifth step (S140) is additionally included. In the fifth step (S140), after receiving the physical quantity to be analyzed, the unit of the physical quantity is determined by interworking with the databases 13 and 15 by the search module 17. It replaces with a zero zone code and converts it to a non-dimensional number, searches for a compatible code corresponding to a non-dimensionalized physical quantity, and identifies a record containing the non-dimensional number by inquiring the compatible code.

At this time, the search module 17, when requesting analysis after inputting a physical quantity to be analyzed through an input window popped up on the terminal screen, the terminal checks whether a unit is included in the physical quantity, and if there is a unit, the unit is zero. After replacing the zone code with a non-dimensional number, searching for a compatible code corresponding to the non-dimensionalized physical quantity, searching for a compatible code to identify the record containing the non-dimensional number, The reference code (equation_address) and mathematical operation method (mathematical_operation) are read and the result is output.

In addition, the first step (S100) further includes a sixth step (S150) of determining whether or not the number is dimensionless by the determination unit 11 of the unit converter 3.

In the sixth step (S150), it is determined whether or not the data fed back by the mining unit 7 is a dimensionless number. If the result of the determination is a dimensionless number, it is directly input to the calculation unit 10, and if it is not a dimensionless number, It is input through the input unit 9.

Claims (15)

A unit converter 3 for converting a physical quantity into a dimensionless number to set a mother number;
An embroidery generator 5 for setting analysis data by generating a child number by converting a parameter set by the unit converter 3 according to a certain rule;
A database (13,15) for storing and fetching the generated parameter and embroidery, and a relational expression matching the parameter and embroidery;
It includes a mining unit 7 interworking with the databases 13 and 15 to retrieve analysis data to perform search, transformation, modeling, evaluation, and prediction,
The mining unit 7 includes a data search unit 20 for searching for nearest data among non-dimensional data stored in a DB according to an input query; A conversion unit 22 for converting data; It includes a modeling unit 24 that selects some of the converted data as target data, and performs association analysis, cluster analysis, classification analysis, and numerical division analysis,
The modeling unit 24 includes a numerical division analysis module 30 that divides and analyzes target data numerically by predetermined intervals,
The numerical division analysis module 30 loads the analysis rule stored in the analysis rule table, divides the analysis target data into a plurality of steps centering on a predetermined number, and the divided plurality of steps is a convergence database and combined analysis history data. Data mining system using dimensionless numbers stored in the base (1). The method of claim 1,
The unit converter 3 includes an input unit 9 for receiving a physical quantity; An operation unit 10 which makes the physical quantity dimensionless by replacing the unit of the input physical quantity with a dimensionless number using a zero zone code; A data mining system (1) using a dimensionless number including an output unit (12) that outputs a dimensionless physical quantity. The method of claim 1,
The analysis data set by the embroidery generator 5 includes a field (mother_number) in which a dimensionless number of a parent relational expression is recorded; A field in which the reference code of the parent relational expression is stored (equation_address); A field containing a mathematical operation method applied to obtain a child relational expression from the parent relational expression (mathematical_operation); And a field (child_number) in which an embroidery obtained through a mathematical operation method for a parameter is recorded. A data mining system using a dimensionless number (1). The method of claim 1,
A search module 17 is additionally included, and the search module 17 interworks with the databases 13 and 15 to convert the unit of the physical quantity into a non-dimensional number by replacing the unit of the physical quantity with a zero zone code after receiving the physical quantity to be analyzed, A data mining system (1) using a dimensionless number, characterized in that a compatible code corresponding to a dimensionless physical quantity is searched, and a record containing the corresponding dimensionless number is identified by searching the compatibility code. The method of claim 4,
The search module 17 enters a physical quantity to be analyzed through an input window popped up on the terminal screen and then requests analysis, the terminal checks whether a unit is included in the physical quantity, and if there is a unit, the unit is a zero zone code. Converted to a non-dimensional number, searched for a compatible code corresponding to a non-dimensionalized physical quantity, and identified a record containing the non-dimensional number by inquiring the compatible code, and then the parameter (mother_number) in the record and the reference code of the relational expression A data mining system using a dimensionless number that reads (equation_address) and mathematical operation method (mathematical_operation) and outputs the result (1). delete The method of claim 1,
The modeling unit 24 includes a correlation analysis module 32 that analyzes a correlation between data; A causal analysis module for analyzing causality between data; A cluster analysis module 36 for analyzing clusters between data; A data mining system using a dimensionless number (1) further comprising an evaluation prediction module 38 that evaluates and predicts the analyzed data. The method of claim 2,
The unit converter 3 further includes a determination unit 11 to determine whether or not the data fed back by the mining unit 7 is a dimensionless number. If the determination result is a dimensionless number, it is directly input to the calculation unit 10 If the number is not a dimensionless number, the data mining system 1 using a dimensionless number inputted through the input unit 9. A first step (S100) of converting a physical quantity into a dimensionless number by the unit converter 3 to set a parameter (Mother number);
A second step (S110) of generating an embroidery (child number) by converting the parameter set by the unit converter 3 according to a predetermined rule by the embroidery generator 5 to set analysis data;
A third step (S120) of storing and retrieving the generated parameter and embroidery, and a relational expression for matching the parameter and embroidery in the databases 13 and 15; And
The mining unit 7 includes a fourth step (S130) of searching, transforming, modeling, evaluating, and predicting by fetching analysis data in connection with the databases 13 and 15,
In the fourth step (S130), the data search unit 20 of the mining unit 7 searches for the nearest data among test data stored in the DB according to the input query (S60); Converting the searched data by the conversion unit 22 (S65); Selecting some of the converted data as target data by the modeling unit 24, performing a correlation analysis, cluster analysis, classification analysis, and numerical division analysis (S70), and evaluation and prediction Including the step (S80),
The numerical division analysis module 30 of the modeling unit 24 loads the analysis rule stored in the analysis rule table, divides the analysis target data into a plurality of steps around a predetermined number, and the divided plurality of steps is convergence data. Data mining method using dimensionless numbers stored in the base and joint analysis history database. The method of claim 9,
In the first step (S100), the input unit 9 of the unit converter 3 receives the physical quantity, and the unit of the physical quantity inputted by the operation unit 10 is replaced with a dimensionless number by a zero zone code to make the physical quantity dimensionless. , Data mining method using a dimensionless number in which the output unit 12 outputs a dimensionless physical quantity. The method of claim 9,
In the second step (S110), the analysis data set by the embroidery generator 5 includes a field (mother_number) in which a dimensionless number of the parent relation is recorded; A field in which the reference code of the parent relational expression is stored (equation_address); A field containing a mathematical operation method applied to obtain a child relational expression from the parent relational expression (mathematical_operation); And a field (child_number) in which an embroidery obtained through a mathematical operation method for a parameter is recorded. The method of claim 9,
A fifth step (S140) is additionally included, and in the fifth step (S140), the unit of the physical quantity is converted into a zero zone code by interworking with the databases 13 and 15 by the search module 17 after receiving the physical quantity to be analyzed. A data mining method using a non-dimensional number, characterized in that it is replaced with a non-dimensional number and converted into a non-dimensional number, a compatible code corresponding to a non-dimensionalized physical quantity is searched, and a record containing the non-dimensional number is identified by searching the compatible code. The method of claim 12,
In the fifth step (S130), the search module 17 inputs a physical quantity to be analyzed through an input window popped up on the terminal screen and then requests analysis, the terminal checks whether a unit is included in the physical quantity, and the unit is If there is, it converts this unit into a zero-zone code and converts it to a non-dimensional number, searches for a compatible code corresponding to the non-dimensionalized physical quantity, and searches the compatibility code to identify the record containing the non-dimensional number, and then the parameter in the record. A data mining method using a dimensionless number that reads (mother_number), a reference code of a relational expression (equation_address), and a mathematical operation method (mathematical_operation) and outputs the result. delete The method of claim 10,
The first step (S100) further includes a sixth step (S150) of determining whether or not a dimensionless number is determined by the determination unit 11 of the unit converter 3, and in the sixth step (S150), the mining unit 7 As a result of determining whether or not the data fed back is a dimensionless number, if it is a dimensionless number, it is directly inputted to the operation unit 10, and if it is not a dimensionless number, the nondimensional number inputted through the input unit 9 is Data mining method used.

Images