KR101844874B1 - System for predicting decision using big data based on data mining techniques and method therefor - Google Patents

Abstract

The present invention discloses a system for predicting a result. More particularly, the present invention relates to a system and method for predicting a result using big data based on a data mining technique with improved accuracy in analyzing data classified into two classes. According to an embodiment of the present invention, learning data is generated by selecting an item for applying raw data according to game contents accumulated from the past in a sports field in which victory and defeat result exists to an SVM algorithm and performing preprocessing, and an analysis model is constructed by applying the selected item to the SVM algorithm, thereby maximizing a hitting ration of the analysis model.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system and method for predicting a result using big data based on data mining techniques,

The present invention relates to a result prediction system, and more particularly, to a system and method for predicting a result using big data based on a data mining technique having improved accuracy in analyzing the divided data into two classes.

In accordance with the rapid development of IT technology, all sports such as soccer, baseball, basketball, volleyball and golf are separated from the passive form in which the spectators simply watch the game played by the athletes, And the function of the TV relay which provided only the information related to the content of the game to the spectator unilaterally was also developed to the level of statistical analysis of the contents of the competition and prediction of the result according to the emergence of the innovative IT technology .

However, conventionally, the information on the prediction of a game against a sport provided on the TV or the Internet is merely a form in which points are given and compared, and the result is provided as a numerical value to be.

This is because a predicted value is calculated using simple input data such as a player or a team record in a sport in which various variables exist, so that there is a limit in that a more accurate and reliable prediction model can not be presented.

In recent years, various types of learning machines based on artificial intelligence technology have appeared, and attempts to apply them to the prediction of a game in a sports game have been gradually increasing. However, until now, only past level data are used as learning data of a learning machine And the result obtained through this is low in accuracy.

Korean Patent Laid-Open Publication No. 10-2012-0136567 (Published date: December 20, 2012) Korean Registered Patent No. 10-1463425 (Registration date: November 13, 2014)

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above problems, and it is an object of the present invention to provide a support vector machine (SVM) algorithm for generating an analysis model having a high hit ratio, There is a problem in providing a system and a method for predicting a result using big data based on a data mining technique to which the present invention is applied.

Also, the present invention provides a system and method for predicting a result using big data based on a data mining technique in which accuracy is improved by performing machine learning by selecting raw data used for generating an analysis model on the basis of time There is a challenge.

In order to solve the above-described problems, a method for predicting a result using big data based on a data mining technique according to an embodiment of the present invention is a method for predicting a result using big data, wherein the preliminary processing of collected big data, Based on a data mining technique that extracts the phenomenon and characteristics of entities included in the data and inputs the query data to the analysis model generated according to the extraction result and provides the partial prediction results classified into at least two classes, Can be used.

The result prediction method includes: collecting raw data including a profile of an entity according to contents of a time-based competition of two teams; Processing the raw data to provide training data corresponding to the profile and including a plurality of features classified into the class; Setting a time range and generating a classifier by mechanically learning learning data in the time range with an SVM algorithm; And predicting a class to which the query data belongs using the classifier.

The step of generating the classifier includes: receiving linear or non-linear processing settings of the learning data; And generating a support vector by receiving the plurality of features as independent variables of the SVM function; And calculating a hyperplane corresponding to the support vector.

Calculating a ratio of the independent variable belonging to each class after predicting the class to which the query data belongs; And applying a weight to each independent variable in proportion to the ratio.

The hyperplane may be defined as a straight line on a second plane that classifies the independent variable into first and second classes.

The hyperplane may be defined as a plane in a cubic space that classifies the independent variable into first and second classes through a kernel trick.

According to another embodiment of the present invention, a result prediction system using big data based on a data mining technique preprocesses the collected big data, and based on the data mining technique, And inputting query data to the analysis model generated according to the extraction result, and using the big data based on the data mining technique for providing the result of the partial prediction divided into at least two classes.

The result prediction system includes: a data collection unit for collecting raw data including a profile of an entity according to contents of time-series games of two teams; A preprocessor for processing the raw data to provide learning data corresponding to the profile and including a plurality of features classified into the class; An SVM unit for setting a time range and learning the learning data in the time range with an SVM algorithm to generate a classifier; And a prediction unit for predicting a class to which the query data belongs using the classifier.

Wherein the SVM unit comprises: a type determination module for receiving linear or nonlinear processing settings of the learning data; A label setting module that sets a classification label corresponding to the class for each of the plurality of features; A vector generation module receiving the profile included in the learning data as an independent variable of the SVM function and generating a support vector; And a hyperplane calculating module for calculating a hyperplane corresponding to the support vector.

Wherein the predicting unit comprises: a rate calculating module for calculating a ratio of the independent variable belonging to each class; And a weight setting module that applies a weight to each independent variable in proportion to the ratio.

The SVM unit may define, for the hyperplane, a straight line on a secondary plane that classifies the independent variable into a first class and a second class.

The SVM unit may define a plane in the third space for classifying the independent variable into the first and second classes through the kernel trick for the hyperplane.

According to the embodiment of the present invention, training data is generated through preprocessing by selecting an item for applying the raw data according to the accumulated game contents to the SVM algorithm in the sports field in which the game result exists, And the analytical model is applied to maximize the hit rate of the analytical model.

Also, according to the embodiment of the present invention, the accuracy of the analysis machine can be improved by performing the machine learning by selectively using the row data according to the time reference.

1 is a diagram illustrating a result prediction method using big data based on a data mining technique according to an embodiment of the present invention.
FIGS. 2A and 2B are diagrams for explaining an SVM algorithm of a result prediction system according to an embodiment of the present invention.
3 is a diagram illustrating a structure of a result prediction system using big data based on a data mining technique according to an embodiment of the present invention.
4A and 4B are tables showing an example of row data used in a result prediction system according to an embodiment of the present invention.

Prior to the description, when an element is referred to as being "comprising" or "including" an element throughout the specification, it is to be understood that the element may be further comprised of other elements . Also, terms such as " unit, "" module, and system," and the like, as used in the specification, designate units that process at least one function or operation , Which may be implemented in hardware, software, or a combination of hardware and software.

Furthermore, the term "embodiment" is used herein to mean serving as an example, instance, or illustration, but the subject matter of the invention is not limited by such example. It is also to be understood that the terms "including, "" having, "and other similar terms are used, but that they do not exclude any additional or different components when used in the claims, Quot; is < / RTI > used in a manner similar to the term " Comprising ".

The various techniques described herein may be implemented with hardware or software, or may be implemented with a combination of both, where appropriate. As used herein, terms such as "unit," "system," and "system" are likewise equally applicable to computer-related entities, And a combination of software, software, or software at the time of execution. Further, in the present invention, both programs and hardware executed in a system including a terminal can be configured in units of modules, and can be recorded in one physical memory, or distributed among two or more memories and recording media.

Hereinafter, a system and method for predicting results using big data based on a data mining technique according to an embodiment of the present invention will be described with reference to the drawings.

1 is a diagram illustrating a result prediction method using big data based on a data mining technique according to an embodiment of the present invention.

Referring to FIG. 1, a result prediction method according to an embodiment of the present invention includes collecting row data including a plurality of items according to time content of two teams (S100), processing raw data, (S110) of preparing learning data including a plurality of features classified into two or more classes, setting a time range and learning corresponding learning data within the time range using a SVM (Support Vector Machine) algorithm Generating a classifier (S120), and estimating a class to which the query data belongs using the classifier (S130).

In the following description, each step constituting the result prediction system using big data based on the data mining technique of the present invention and each constituent part constituting it are performed without any other instruction.

Big data is defined in terms of many vast amounts of form, numbers, text, and images that can not be scaled as digital technology evolves. Such big data is characterized by a volume of data, a velocity of data generation, and a variety of types as a difference from existing data.

The amount of the data generally refers to a data attribute of a size of more than tens of terabytes (Tera Bytes) or dozens of petabytes (Peta Bytes), but the present invention is not limited thereto. The generation rate is a property that can process and analyze a large amount of data quickly Lt; / RTI > In other words, in the future IT environment, various kinds of data are produced at a very high speed, and this means performance capable of storing, distributing, collecting and analyzing in real time.

Also, the variety of forms means various types of data, and can be classified into stereotyped, semi-stereotyped, and non-stereotyped data according to the type of stereotyping as an example.

In the embodiment of the present invention, a system and a method for predicting the results of various fields using such big data are proposed, and a data mining technique is used for this purpose.

Data mining is defined as the process of discovering the correlation between entities contained in the above-mentioned big data, extracting useful information, and utilizing it for decision making.

Such data mining can be classified into Classification, Predication, Estimation, Clustering, and Affinity Grouping. The classification is performed by observing new characteristics of entities in the data, It belongs to a defined group, and prediction is to find future actions or values that do not yet occur.

In addition, the estimation is to find successive results, not to deal with bipartite and discrete results, such as classification; clustering is to subdivide heterogeneous groups into several homogeneous subgroups; .

In the embodiment of the present invention, a system and method for generating a classifier and predicting a result through a data mining technique using game information on a sports game accumulated from the past to the present are used as big data.

Specifically, the step of collecting row data including a plurality of items according to the contents of the hourly games of the two teams collects raw data to be an object of learning to build an analysis model, The data on the game content is input as raw data to a sports game such as a soccer game played by the team, and the input data is collected in the result prediction system.

Here, the row data to be subjected to the machine learning may be data including a plurality of items related to the game contents of the sports field to be predicted, the items may be selected by the system operator, And data of the second matrix may be used as fields of the game content.

Next, the step of preparing learning data including a plurality of features corresponding to the items and processing the low data and classified into two or more classes (S110) may include applying the collected raw data to the SVM algorithm It is a step of machining into a form that can be made.

As described above, the row data may be in the form of a second-order matrix having numerical values for a plurality of items. Step S110 is a step of preprocessing the data. In step S110, an item-specific value is set for each identification code so that it can be substituted into the SVM algorithm Into learning data in the form of having.

The step of setting a time range and learning the corresponding learning data in the time range using the SVM algorithm (S120) includes a step S120 of applying the SVM algorithm to the learning data prepared according to the user's intention, , Thereby generating a classifier capable of obtaining the differentiated result.

Here, the SVM algorithm is mainly used when the data to be handled is classified into two classes. In the present invention, data on the game contents of two teams in past sports games, for example, If a model for victory and defeat is created, it is possible to determine which of the items of data has a high influence on the game, and a new game can be predicted by inputting new data.

That is, in the embodiment of the present invention, a player is applied as an entity included in the data, and at least one of a plurality of profiles for each player according to a competition is used as learning data.

As an example, as a result of a plurality of behaviors generated by an individual in a football game of the past two teams, and as a result thereof, the profile includes a player's goals, assists, effective shots, other shots, block shots, offside, It indicates items such as the number of fouls, the number of yellow cards received, the number of red cards received, the success rate of pass, and the share. Using this as learning data, a classifier classified into two classes, game and defeat, is created through machine learning based on the SVM algorithm Then, if any one of the above items is entered into the classifier for the subsequent game, it is possible to predict to which class the result according to the item belongs, namely, victory or defeat.

The subdivision of step S120 may include receiving linear or nonlinear processing settings of learning data, generating a support vector by receiving a plurality of features as independent variables of the SVM function, and calculating a hyperplane corresponding to the support vector And the technical idea of the present invention will be described by the mathematical formula derived for performing each step below.

More specifically, the SVM algorithm finds a hyperplane based on a straight line (y) in which a plurality of samples '?' And '?' Are divided into first and second classes, respectively, (1) " (1) "

Here, [omega] is a normal vector perpendicular to the straight line, the size is [omega] [theta], and it can have a property of rotating a straight line. B is a scalar constant, and x is an arbitrary data vector.

Next, based on Equation ( ₁ ), if the two straight lines (y) which respectively meet the first data (x ₁ , x ₂ ) of the two classes are 1, -1,

Here, x ₁ and x ₂ are referred to as a support vector. At this time, the margin (M) between the two straight lines becomes the ω direction component of the support vector.

Therefore, the margin M satisfies the following equation (3) as shown in the following equation (3).

로 바꿀 수 있다.Here, in order to maximize the margin M, it can be seen that < RTI ID = 0.0 >

.

All data (x _i ) belonging to the two classes must satisfy the following expression (4) as a conditional expression.

Accordingly, if y _i ≥1 if any data x _i1 is in the first class and y _i ≤-1 if any data x _i2 is in the second class, (5) " (5) "

에 대한 최적화 문제는 수학식 6의 최적화 문제로 변환할 수 있다.Under the condition of Equation (5) above,

Can be transformed into the optimization problem of Equation (6).

Here, α _i is a Lagrange multiplier, and α _i ≥0 in the optimization problem in which the minimum value is obtained, and Equation (7), which is an expression for ω, when the partial differential is performed.

Substituting Equation (7) into Equation (6), it is as follows.

On the other hand, the equation (9) can be expressed by a matrix H used for learning data.

Accordingly, when Equation (8) is converted into Equation (8) to obtain the minimum value of? _I , and Equation (9) is substituted, it can be converted into Equation (10) below, .

Further, based on Equation (5), b can be calculated by converting to Equation (11), which is an equation relating to b.

Here, by calculating? And b according to the above-described Equations 1 to 11, it is possible to calculate a hyperplane and generate a classifier.

On the other hand, the SVM algorithm described above is a hard margin method for calculating a boundary equation that strictly separates two classes. When the amount of learning data increases, as shown in FIG. 2B, the hyperplane of the input space is not linear It becomes a nonlinear shape and the noise generation rate is increased. In order to improve these problems, a soft margin method has been proposed in which classes are classified by converting the characteristic space into a tertiary plane by raising the dimension.

More specifically, it transforms into a higher-dimensional space such that linear separation is possible as shown in Equation (12).

Then, a conversion function that is easy to perform scalar multiplication is selected. An example of such a transform function is shown in Equation (13).

The matrix H of the learning data is calculated through Equations (12) and (13) as follows.

In the above Equation (14), a set of scalar multiplications among conversion functions is called a kernel. The kernel can easily calculate the matrix H by transforming the distribution which is difficult to linearly classify into a space with one dimension that is linearly classifiable. Such kernels include polynomial kernels and hyperbolic tangent kernels. Therefore, the hyperplane is defined as a plane shape rather than a straight line through a kernel trick by a kernel function.

When the classifier is generated according to the above-described procedure, the class is determined by inputting query data to be predicted thereafter.

Referring back to FIG. 1, in step S130 of predicting the class to which the query data belongs using the classifier, the query data to be predicted is input to the classifier, and the classification result, that is, And outputting the prediction data.

In particular, the query data can be composed of one or more items of the same kind as the learning data. When the items are input, the game can be predicted according to the classification of the classes belonging to the corresponding team according to the input items.

According to an embodiment of the present invention, there is provided a method for estimating a class of a query data, the method comprising: calculating a ratio between the independent variables belonging to each class, A step of applying weights may be performed to further confirm the accuracy of the current analysis model.

Hereinafter, a result prediction system using big data based on a data mining technique according to an embodiment of the present invention will be described with reference to the drawings.

In the following description, the result prediction system is implemented in a computing device, and its components and the modules included therein may be implemented in a program form written in a predetermined programming language and recorded in a recording medium.

3 is a diagram illustrating a structure of a result prediction system using big data based on a data mining technique according to an embodiment of the present invention.

3, the result prediction system 100 of the present invention includes a data collecting unit 110 for collecting raw data including a plurality of items according to contents of time-series games of two teams, A preprocessing unit 120 for providing learning data including a plurality of features corresponding to at least two classes, a SVM for setting a time range and learning the corresponding learning data in the time range using an SVM algorithm, And a prediction unit 140 for predicting a class to which the query data belongs using the classifier.

The data collecting unit 110 collects raw data to be a learning target for building an analysis model. The result prediction system of the present invention is to provide an analysis model for sports events having differentiated results such as victory or defeat, and collects a plurality of row data selected by the system operator.

Here, the raw data to be subjected to the machine learning may be data including a plurality of items related to the contents of a game in the sports field to be predicted, In the case of a soccer game, the above items include the goals, assists, effective shots, other shots, block shots, offside, number of fouls, number of fouls, number of yellow cards received, The number of times of receipt, the pass success rate, and the share of the pass.

In particular, the row data may be data of a secondary matrix type having an item name as an identification code and an item of a game content as a field, and the preprocessing unit 120 prepares a field value of each ID code on the secondary matrix, As shown in FIG.

That is, the preprocessing unit 120 processes the collected raw data into a form applicable to the SVM algorithm. The row data has a numeric value for a plurality of items and is converted into learning data of a form having a value for each item for each identification code, and is input to the SVM unit 130 as the value of the independent variable.

The SVM unit 130 generates an analysis module corresponding to the learning data provided from the preprocessing unit 120.

That is, the SVM unit 130 is a learning machine based on a support vector machine algorithm, and generates a classifier including a hyperplane for classifying win / lose according to the value of each item included in the learning data.

Here, the numerical value included in the learning data is input to a variable declared as an integer (int), a float, or the like in the learning machine according to the type, and the hyperplane calculated according to the above-described Equations 1 to 14 The class is divided into one class.

In order to implement this function, the SVM unit 130 includes a type determination module 121 that receives linear or nonlinear processing settings of learning data, a label setting module 121 that sets a classification label corresponding to the class for each of a plurality of features A vector generation module 123 that receives the training data as independent variables of the SVM function and generates a support vector, and a hyperplane calculating module 124 that calculates a hyperplane corresponding to the support vector.

In particular, the type determination module 121 receives a type such as linear or non-linear for selecting an optimal SVM algorithm according to the amount of input learning data, and an SVM algorithm for creating a classifier corresponding to the type is determined .

When the SVM algorithm is determined, the vector generation module 132 generates the support vector according to Equations 1 to 14 described above, and the hyperplane calculating module 133 generates the classifier by calculating the hyperplane.

The predicting unit 140 inputs the query data (query_data) to be predicted to the classifier and calculates which class of the classified result, namely, victory or defeat, and outputs the predicted data (predict_data).

Here, the query data (query_data) may be composed of one or more items of the same kind as the learning data. When the item is input, the predicted data (predict_data) As a result, the game can be predicted.

Although not shown, it is possible to calculate the accuracy of the classifier generated by comparing the predicted data (predict_data) after the prediction with the actual game result.

In order to realize this function, the prediction unit 140 includes a ratio calculation module 141 for calculating a ratio of independent variables belonging to each class, a weight setting module 142 for applying a weight to each independent variable in proportion to the ratio, . ≪ / RTI >

Hereinafter, a method of deriving learning data in the present invention will be described with reference to an example of a row data type used in a result prediction system using big data based on a data mining technique according to an embodiment of the present invention, with reference to the drawings.

4A and 4B are tables showing an example of row data used in a result prediction system using big data based on a data mining technique according to an embodiment of the present invention.

Referring to FIGS. 4A and 4B, the result prediction system of the present invention is a system for predicting the performance of a player who occurs in time according to progress of a game, such as a football game, through learning data using a predetermined item Perform machine learning and generate classifiers.

That is, the result prediction system according to the embodiment of the present invention generates an analysis model by using at least one of a plurality of profiles for each player according to a competition as a factor of learning data.

FIGS. 4A and 4B show data 200 and 210 corresponding to the game contents of the players of the two teams, which occurred in the soccer game between 'Everton' and 'Liverpool' which occurred in the past, As shown in FIG.

Each item 202 included in the game content includes a goal G, an assist A, an effective shot GA, another shooting SG, a block shoot BS, an offside O, a foul count FC, , The number of fouls (FS), the number of yellow cards received (YC), the number of red cards received (RC), the pass success rate (PS), and the occupancy rate (TP) Is input as an independent variable of the learning data.

Particularly, the above-described data can collect more data at a previous time as data up to a specific time point after the game starts, input all the raw data for each time into the preprocessor 120 according to the intention of the system operator, The analysis model is constructed.

In addition, the illustrated low data 200 and 210 are games in which 'Everton' has won 1: 0 with 'Liverpool' at the end of the game, and the score of 'Mane S.' in the 'Liverpool' ) Item is deterministic. According to the raw data 200, items having 'Mane S.' as an identification code have the greatest influence on victory over items of identification code, and these items are analyzed After the generation of the model, the prediction performance can be further improved by calculating the ratios and setting the weights for the corresponding player or item when inputting the query data for prediction.

While a great many are described in the foregoing description, it should be construed as an example of preferred embodiments rather than limiting the scope of the invention. Accordingly, the invention is not to be determined by the embodiments described, but should be determined by equivalents to the claims and the appended claims.

100: result prediction system 110: data collection unit
120: preprocessing unit 130: SVM unit
131: type determination module 132: vector generation module
133: hyperplane calculating module 140: predictor
141: Rate calculation module 142: Weight setting module

Claims (12)

delete delete delete delete delete delete Extracting phenomena and characteristics of entities included in the learning data based on the data mining technique, inputting query data into the analysis model generated according to the extraction result, As a result prediction system using big data based on a data mining technique that provides partial prediction results,
A data collecting unit for collecting, as raw data, big data including a profile of an individual according to the contents of time-series games of the two teams; A preprocessor for processing the raw data to provide learning data corresponding to the profile and including a plurality of features classified into the class; An SVM unit for setting a time range and learning the learning data in the time range with an SVM algorithm to generate a classifier; And a predictor for predicting a class to which the query data belongs using the classifier, based on the data mining technique. 8. The method of claim 7,
The game content is a game content of a soccer game, the object is a name of a player, and the profile includes a player's goal, assist, effective shooting, other shooting, block shoot, offsite, number of fouls, number of fouls, , The number of times the red card was received, the pass success rate, and the share of the player. 8. The method of claim 7,
The SVM unit,
A type determination module that receives linear or nonlinear processing settings of the learning data;
A label setting module that sets a classification label corresponding to the class for each of the plurality of features;
A vector generation module receiving the profile included in the learning data as an independent variable of the SVM function and generating a support vector; And
A hyperplane calculating module for calculating a hyperplane corresponding to the support vector
Based on the data mining technique. 10. The method of claim 9,
The predicting unit,
A ratio calculating module for calculating a ratio of the independent variable belonging to each class; And
A weight setting module that applies a weight to each independent variable in proportion to the ratio
Based on the data mining technique. 10. The method of claim 9,
The SVM unit,
A result prediction system using big data based on a data mining technique in which the hyperplane is defined as a straight line on a secondary plane that classifies the independent variable into first and second classes. 10. The method of claim 9,
The SVM unit
Wherein the independent variable is defined as a plane in a cubic space that classifies the independent variable into a first class and a second class through the kernel trick for the hyperplane.

Images