KR102205102B1 - Prediction of number of bitcoin network transactions based on machine learning - Google Patents

Abstract

Disclosed is a technology for predicting the number of bitcoin network transactions based on machine learning. According to one embodiment, the method for predicting the number of transactions comprises the steps of: constructing a machine learning model for predicting the number of transactions using virtual currency data; performing a pre-processing process on the virtual currency data to select learning data to be inputted to the constructed machine learning model; and predicting the number of transactions in virtual currency as the selected learning data is inputted to the constructed machine learning model to be learned.

Description

Machine Learning-based Bitcoin Network Transaction Count {PREDICTION OF NUMBER OF BITCOIN NETWORK TRANSACTIONS BASED ON MACHINE LEARNING}

The description below relates to a method and system for predicting the number of Bitcoin transactions.

The number of bitcoin transactions has increased dramatically over the past few years. As the number of transactions increased, the Bitcoin network became active. However, side effects are occurring due to the increase in the number of bitcoin transactions. The transaction fee has increased, but the transaction confirmation time has been delayed. It can be seen that the number of bitcoin transactions and the bitcoin price show a clear positive linear relationship, and the number of bitcoin transactions is a factor affecting the bitcoin price. Predicting the number of bitcoin transactions is important to prepare for side effects that occur on the bitcoin network.

For example, Korean Patent Laid-Open No. 10-2018-0014534 relates to a blockchain-based transaction verification system and its method. Whether a node receives a block generated by a collection of transactions, and whether the node is assigned block verification authority for the block. When it is determined that the block verification authority is granted, a configuration for verifying the transaction of the received block is initiated. However, the conventional technique only verifies the transaction and cannot predict the number of transactions.

Accordingly, in order to prepare for the growth and side effects of the Bitcoin network, a technique for predicting the number of Bitcoin transactions is required.

A method and system for designing a machine learning model that predicts the number of Bitcoin transactions can be provided.

The method of predicting the number of transactions may include constructing a machine learning model for predicting the number of transactions using virtual currency data; Selecting training data to be input to the configured machine learning model by performing a preprocessing process on the virtual currency data; And predicting the number of virtual currency transactions by inputting and learning the selected training data into the configured machine learning model.

The selecting of the learning data includes collecting virtual currency data including block data and transaction data included in the virtual currency block, and performing statistical processing of the block data and transaction data included in the collected virtual currency data. It may include obtaining currency block data and transaction statistics data.

The step of selecting the learning data includes selecting learning data by applying a correlation analysis to the obtained virtual currency block data and transaction statistics data, and any of Pearson correlation analysis or Spearman correlation analysis as the correlation analysis. One can be applied.

The step of selecting the training data includes selecting a training data (train set) by applying a correlation analysis between the virtual currency block data and transaction statistics data of the Nth block and the number of N+1th virtual currency transactions, and the It may include the step of classifying the data that has not been selected as the training data as validation data (validation set).

The predicting of the number of transactions includes inputting and training the selected training data based on a linear regression model and an LSTM model existing in the configured machine learning model, respectively, and the linear regression model includes input data As input data, cryptocurrency block data and transaction statistics are used, and the LSTM model can use cryptocurrency block data, transaction statistics data, and the number of transactions as input data.

The predicting of the number of transactions includes the training data, the number of networks of the machine learning model, a loss function that evaluates the performance of the machine learning model, and an optimization function that is a function for reducing the loss function. ), a sequence length that is the length of the training data, and a hyper that configures the machine learning model by setting at least one element of the number of hidden units existing in the cell of the machine learning model. It may include the step of optimizing the parameter (Hyper-parameter).

The transaction number prediction system includes: a model construction unit that constructs a machine learning model for predicting the number of transactions using virtual currency data; A selection unit selecting training data for input into the configured machine learning model by performing a preprocessing process on the virtual currency data; And a prediction unit that predicts the number of transactions of virtual currency by inputting and learning the selected training data in the configured machine learning model.

The selection unit collects cryptocurrency data including block data and transaction data included in the cryptocurrency block, and performs statistical processing of block data and transaction data included in the collected cryptocurrency data to provide cryptocurrency block data and transactions. Statistical data can be obtained.

The selection unit may include selecting learning data by applying a correlation analysis to the acquired virtual currency block data and transaction statistics data, and may apply either Pearson correlation analysis or Spearman correlation analysis as the correlation analysis. .

The selection unit selects a training data (train set) by applying a correlation analysis between the virtual currency block data and transaction statistics data of the Nth block and the number of N+1th virtual currency transactions, and is not selected as the training data. Data can be classified as validation set.

The prediction unit includes inputting and training the selected training data based on a linear regression model and an LSTM model existing in the configured machine learning model, respectively, and the linear regression model includes virtual currency block data and The transaction statistics data is used, and the LSTM model can use virtual currency block data, transaction statistics data, and the number of transactions as input data.

The prediction unit may include the training data, the number of networks of the machine learning model, a loss function for evaluating the performance of the machine learning model, an optimization function that is a function for reducing the loss function, and the training data. Hyper-parameter constituting the machine learning model by setting at least one element of the sequence length, which is the length of the machine learning model, and the number of hidden units in the cell of the machine learning model. ) Can be optimized.

The number of cryptocurrency transactions that are important for maintaining and growing the cryptocurrency network can be predicted by applying a machine learning model.

1 is a diagram illustrating an operation of a system for predicting the number of transactions according to an embodiment.
2 is a block diagram illustrating a configuration of a system for predicting the number of transactions according to an embodiment.
3 is a flowchart illustrating a method of predicting the number of transactions in the system for predicting the number of transactions according to an embodiment.
4 and 5 are examples of comparing performance of a machine learning model to which correlation analysis is applied in a system for predicting the number of transactions according to an embodiment.
6 and 7 are examples for explaining optimization of a hyper parameter constituting a machine learning model in a system for predicting the number of transactions according to an embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

Virtual currency (cryptocurrency) is a type of electronic money that remains in the form of information on a computer, etc., and is traded only on cyber without real objects, and is processed by a distributed system processing method using blockchain technology, and various types of virtual currency can exist. have. In the embodiment, for convenience of explanation, bitcoin as a virtual currency will be described as an example.

1 is a diagram illustrating an operation of a system for predicting the number of transactions according to an embodiment.

The transaction number prediction system may perform a preprocessing process 110 and a machine learning model design process 120. In the preprocessing process 110, the system for predicting the number of transactions may collect block data and transaction data included in the bitcoin block from the bitcoin core. At this time, CentOS operating system can be used for collection, and MongoDB can be used for database software. There can be various types of collected bitcoin block data and transaction data. For example, Bitcoin block data and transaction data type, including the number of transactions included in the block, the weight of the block, the size of the block, the virtual size of the block, the number of inputs in the transaction, and the transaction. There may be the number of outputs being performed, the transaction amount of the transaction, the transaction fee, the size of the transaction, the virtual size of the transaction, the input value of the transaction, and the output value of the transaction.

The transaction number prediction system may perform statistical processing to obtain statistical information from the collected bitcoin block data and statistical data. The system for predicting the number of transactions can apply statistical processing to each transaction unit. At this time, the system for predicting the number of transactions may derive sum, product, maximum value, minimum value, and average value through statistical processing. For example, statistical processing of a preset number of times (eg, 1, 2, etc.) may be performed on transaction data. It may be composed of a plurality of types (eg, 84 types) of bitcoin block data and transaction statistics data that have performed such statistical processing.

The transaction number prediction system may select learning data by applying correlation analysis to the virtual currency block data and transaction statistics data obtained as statistical processing is performed. In this case, by applying the correlation analysis, it may be classified into training data input to the machine learning model and verification data not used for training. The training data can be used to train the machine learning model, and the verification data can be used to verify the performance of the machine learning model.

Specifically, correlation analysis is to analyze the presence or absence of a linear relationship between two variables. The two variables may be independent or correlated. At this time, the strength of the relationship between the two variables is called correlation. In an embodiment, correlation analysis may be used to select training data suitable for a machine learning model. For example, Pearson correlation analysis or Spearman correlation analysis can be used.

First, a method of selecting training data suitable for a machine learning model using Pearson's correlation analysis will be described. Pearson correlation analysis is performed using Pearson correlation coefficient. At this time, the Pearson correlation coefficient is a quantification of the linear correlation between two variables. The Pearson correlation coefficient has a value between -1 and +1, and a value can be derived by Equation 1.

Equation 1:

는 블록 통계 데이터의 표본, S _B 는 블록 통계 데이터의 표준 편차, S _T 는 트랜잭션 수의 표준 편차를 의미한다. 또한, 피어슨 상관계수 p의 범위에 따라 다음과 같은 해석을 할 수 있다. 피어슨 상관계수의 값이 -1인 경우 완벽한 음의 선형적 관계, +1인 경우 완벽한 선형적 관계를 의미한다. p is the Pearson correlation coefficient, B is the block statistics, T is the number of transactions, n is the number of data, S is the standard deviation, and bar is the sample mean. Specifically, B _i is block statistical data,

Is the sample of block statistics data, S _B is the standard deviation of block statistics data, and S _T is the standard deviation of the number of transactions. In addition, according to the range of the Pearson correlation coefficient p , the following interpretation can do. If the value of Pearson's correlation coefficient is -1, it means a perfect negative linear relationship, and when it is +1, it means a perfect linear relationship.

Next, a method of selecting training data suitable for a machine learning model using Spearman correlation analysis will be described. Spearman correlation analysis is performed using the Spearman correlation coefficient. The Spearman correlation coefficient is a correlation coefficient in the case of using a ranking instead of a value of a variable, and values can be derived using the ranking after changing the order of ranking by ranking values from smallest ones. The value of the Spearman correlation coefficient may be derived by Equation 2.

Equation 2:

는 블록 통계 데이터의 표본 평균,

는 트랜잭션 수의 표본 평균을 의미한다. 또한, 스피어만 상관계수 p의 범위에 따라 다음과 같은 해석을 할 수 있다. 스피어만 상관계수의 값은 -1내지 +1사이의 값을 갖는다. -1인 경우 두 변수의 순위가 완전 반대이고 +1인 경우 두 변수의 순위가 완벽히 일치함을 의미한다. p is a Spearman correlation coefficient, n is the number of data, B _i is a rank of the i-th data block in the data statistics, T _i is a rank, bar of the i-th data in the number of transactions is the sample mean.

Is the sample mean of the block statistical data,

Is the sample average of the number of transactions. In addition, according to the range of the Spearman correlation coefficient p , the following interpretation can do. The values of the Spearman correlation coefficient range from -1 to +1. A value of -1 means that the ranks of the two variables are completely opposite, and a value of +1 means that the ranks of the two variables are perfectly matched.

In order to predict the number of transactions, the system for predicting the number of transactions performs a correlation analysis between the number of bitcoin block data and transaction statistics data of the Nth block and the number of N+1th bitcoin transactions, and then calculates the correlation coefficient. They can be listed in ascending order by size. Referring to FIGS. 4 and 5, it is an example of comparing the performance of a machine learning model to which correlation analysis is applied in a transaction number prediction system. In the case of Pearson correlation analysis, in order to compare the performance of the machine learning model according to the training data, the transaction number prediction system may select data having a positive linear relationship and data having no linear relationship. For example, three data having a positive linear relationship and three data having no linear relationship may be selected. At this time, there is no data indicating a negative linear relationship with the number of transactions.

In the case of Spearman correlation analysis, in order to compare the performance of the machine learning model according to the training data, the transaction number prediction system may select data close to -1 or +1 and data that are not correlated. For example, you can select 4 data close to -1 or +1 and 3 data that are not correlated. Referring to FIGS. 4 and 5, it can be seen that the items of the top 3 numbers having a large correlation coefficient value through two correlation analysis are matched by sum.nVout, sum.vsize, and sum.nVin. However, in the case of Pearson correlation analysis, the correlation coefficient value of sum.nVin is the largest, and in the case of Spearman correlation analysis, the correlation coefficient value of sum.vsize is the largest.

The transaction number prediction system can select a machine learning model. The transaction number prediction system can select two machine learning models. In this case, the transaction number prediction system may select a machine learning model including a linear regression model to which the simplest algorithm among machine learning algorithms is applied and an LSTM model specialized for time series data prediction. The transaction number prediction system can compare the performance of the linear regression model and the LSTM model using the same training data, and can perform the performance comparison of each model on the data item selected through correlation analysis. In the case of the linear regression model, the number of bitcoin transactions can be predicted using bitcoin block data and transaction statistics data as input data when training, and in the case of the LSTM model, bitcoin block data and transaction statistics as input data when learning. The number of transactions can be predicted using the number of data and transactions as input data.

In constructing a machine learning model, the transaction number prediction system can optimize elements for setting the machine learning model. The hyper parameter may mean an element that can be arbitrarily set in constructing a machine learning model, and by adjusting these elements, a machine learning model with optimal performance can be derived. The system for predicting the number of transactions is the training data, the number of networks of the machine learning model, the loss function that evaluates the performance of the machine learning model, the optimize function, which is a function to reduce the loss function, and the length of the training data. Optimizing hyper-parameters constituting the machine learning model by setting at least one factor among the sequence length and the number of hidden units in the cell of the machine learning model can do. For example, in the case of a linear regression model, hyperparameters including the number of neural networks, a loss function that evaluates the performance of the model, and an optimization function that is a function to reduce the loss function Can be set. In the case of an LSTM model, the number of neural networks, the loss function to evaluate the performance of the model, the optimize function to reduce the loss function, and the length of data to be used for one-time training It is possible to set a hyper parameter including the sequence length and the number of hidden units existing in the LSTM cell. In order to evaluate the performance of the trained machine learning model, the hyperparameter can be optimized by analyzing the validation data and the loss function applied to the training data.

6 and 7 are examples for explaining optimization of hyper parameters constituting a machine learning model. The hyperparameter of the machine learning model can be optimized by using sum.nVin, which has the best performance in predicting the number of bitcoin transactions according to FIGS. 4 and 5, as training data. As an example, referring to FIG. 6, it is an example showing that the number of hidden units is optimized. The system for predicting the number of transactions may set the sequence length to 1 to derive the optimal number of hidden units while increasing the number of hidden units. At this time, it can be seen that even if the number of hidden units exceeds 8, the MSE does not decrease. Accordingly, the transaction number prediction system may set the number of hidden units to 8. As another example, referring to FIG. 7, it is an example showing that the sequence length is optimized. The transaction number prediction system sets the number of hidden units to 8, and increases the sequence length to derive the optimal sequence length. At this time, it can be seen that even if the number of sequence lengths exceeds 32, the MSE does not decrease. Accordingly, the system for predicting the number of transactions may set the sequence length to 32.

As the transaction number prediction system according to an embodiment applies Pearson correlation analysis as a methodology for selecting training data of an LSTM model, it can be seen that the performance of the LSTM model improves as the value of the Pearson correlation coefficient increases. In addition, optimization of the hyper parameters constituting the LSTM model can be applied through experiments.

The system for predicting the number of transactions according to an embodiment may search for a configuration representing the minimum value through verification of a machine learning model and an experimental MSE. In addition, the transaction number prediction system can find that the value of the sequence length rather than the number of hidden units is a hyper parameter that affects the performance of the LSTM model.

In addition, the MSE change of the LSTM model is large according to the change of the sequence length value under the same condition.

2 is a block diagram illustrating the configuration of a system for predicting the number of transactions according to an embodiment, and FIG. 3 is a flowchart illustrating a method of predicting the number of transactions in the system for predicting the number of transactions according to an embodiment.

The processor of the transaction number prediction system 100 may include a model construction unit 210, a selection unit 220, and a prediction unit 230. Components of such a processor may be expressions of different functions performed by the processor according to a control command provided by a program code stored in the transaction number prediction system 100. The processor and components of the processor may control the transaction number prediction system 100 to perform steps 310 to 330 included in the method for predicting the number of transactions of FIG. 3. In this case, the processor and the components of the processor may be implemented to execute an instruction according to the code of the operating system included in the memory and the code of at least one program.

The processor may load program code stored in a program file for a method of predicting the number of transactions into the memory. For example, when a program is executed in the transaction number prediction system 100, the processor may control the transaction number prediction system 100 to load the program code from the program file into the memory under the control of the operating system. At this time, each of the processor and the model construction unit 210, the selection unit 220, and the prediction unit 230 included in the processor executes a command of a corresponding part of the program code loaded into the memory, and subsequent steps 310 to 330 ) May be different functional representations of the processor to execute.

In step 310, the model construction unit 210 may construct a machine learning model for predicting the number of transactions using virtual currency data.

In step 320, the selection unit 220 may select training data for input into the configured machine learning model by performing a preprocessing process on the virtual currency data. The selection unit 220 collects cryptocurrency data including block data and transaction data included in the cryptocurrency block, and performs statistical processing of block data and transaction data included in the collected cryptocurrency data to provide cryptocurrency data and transactions. Statistical data can be obtained. The selection unit 220 may select learning data by applying correlation analysis to the acquired virtual currency block data and transaction statistics data. The selection unit 220 may apply either Pearson correlation analysis or Spearman correlation analysis as correlation analysis. The selection unit 220 selects a training data (train set) by applying a correlation analysis between the virtual currency block data and transaction statistics data of the Nth block and the number of N+1th virtual currency transactions, and is not selected as the training data. The obtained data can be classified as a validation set.

In step 330, the prediction unit 230 may predict the number of transactions of the virtual currency by inputting and learning the selected training data in the configured machine learning model. The prediction unit 230 may input and train training data selected based on a linear regression model and an LSTM model existing in the configured machine learning model, respectively. The prediction unit 230 includes training data, the number of networks of the machine learning model, a loss function that evaluates the performance of the machine learning model, an optimization function that is a function for reducing the loss function, and the length of the training data. Optimizing hyper-parameters constituting the machine learning model by setting at least one factor among the sequence length and the number of hidden units in the cell of the machine learning model can do.

The apparatus described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments are, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA). , A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions, such as one or more general purpose computers or special purpose computers. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. Can be embodyed in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (12)

In the method of predicting the number of transactions,
Constructing a machine learning model for predicting the number of transactions using virtual currency data;
Selecting training data to be input to the configured machine learning model by performing a preprocessing process on the virtual currency data; And
Predicting the number of virtual currency transactions by inputting and learning the selected training data into the configured machine learning model
How to predict the number of transactions, including. The method of claim 1,
The step of selecting the training data,
Collects virtual currency data including block data and transaction data included in the virtual currency block, and acquires virtual currency block data and transaction statistics data through statistical processing of the block data and transaction data included in the collected virtual currency data Steps to
How to predict the number of transactions, including. The method of claim 2,
The step of selecting the training data,
Selecting learning data by applying correlation analysis to the obtained virtual currency block data and transaction statistics data
Including,
Applying either Pearson correlation analysis or Spearman correlation analysis as the correlation analysis
How to estimate the number of transactions. The method of claim 3,
The step of selecting the training data,
The training data (train set) is selected by applying correlation analysis between the cryptocurrency block data and transaction statistics data of the Nth block and the number of cryptocurrency transactions of the N+1th, and the data that has not been selected as the training data is verified. Classification as (Validation set)
How to predict the number of transactions, including. The method of claim 1,
Predicting the number of transactions,
Inputting and learning the selected training data based on a linear regression model and an LSTM model existing in the configured machine learning model
Including,
The linear regression model uses virtual currency block data and transaction statistics as input data,
The LSTM model uses cryptocurrency block data, transaction statistics, and the number of transactions as input data.
How to estimate the number of transactions. The method of claim 5,
Predicting the number of transactions,
The training data, the number of networks of the machine learning model, a loss function for evaluating the performance of the machine learning model, an optimization function that is a function for reducing the loss function, a sequence that is the length of the training data Optimizing hyper-parameters constituting the machine learning model by setting at least one factor among the length (Sequence Length) and the number of hidden units in the cell of the machine learning model. step
How to predict the number of transactions, including. In the transaction number prediction system,
A model construction unit configuring a machine learning model for predicting the number of transactions using virtual currency data;
A selection unit selecting training data for input into the configured machine learning model by performing a preprocessing process on the virtual currency data; And
A prediction unit that predicts the number of transactions of virtual currency by inputting and learning the selected training data into the configured machine learning model
Transaction number prediction system comprising a. The method of claim 7,
The selection unit,
Collects virtual currency data including block data and transaction data included in the virtual currency block, and acquires virtual currency block data and transaction statistics data through statistical processing of the block data and transaction data included in the collected virtual currency data doing
A system for predicting the number of transactions, characterized in that. The method of claim 8,
The selection unit,
Including selecting learning data by applying a correlation analysis to the obtained virtual currency block data and transaction statistics data,
Applying either Pearson correlation analysis or Spearman correlation analysis as the correlation analysis
A system for predicting the number of transactions, characterized in that. The method of claim 9,
The selection unit,
The training data (train set) is selected by applying correlation analysis between the cryptocurrency block data and transaction statistics data of the Nth block and the number of cryptocurrency transactions of the N+1th, and the data that has not been selected as the training data is verified. Classified as (Validation set)
A system for predicting the number of transactions, characterized in that. The method of claim 7,
The prediction unit,
Including each inputting and training the selected training data based on a linear regression model and an LSTM model existing in the configured machine learning model,
The linear regression model uses virtual currency block data and transaction statistics as input data,
The LSTM model uses cryptocurrency block data, transaction statistics, and the number of transactions as input data.
A system for predicting the number of transactions, characterized in that. The method of claim 11,
The prediction unit,
The training data, the number of networks of the machine learning model, a loss function for evaluating the performance of the machine learning model, an optimization function that is a function for reducing the loss function, a sequence that is the length of the training data Optimizing hyper-parameters constituting the machine learning model by setting at least one factor among the length (Sequence Length) and the number of hidden units in the cell of the machine learning model.
A system for predicting the number of transactions, characterized in that.

Images