KR20200063361A - A feature level fusion system and method for stock price forecasting - Google Patents

Abstract

Disclosed are a feature convergence based stock price prediction method and a system therefor. According to an embodiment of the present invention, a stock price prediction method performed by a prediction system comprises the steps of: extracting each feature by learning KOSPI index data and foreign index data by using a stacked noise elimination auto-encoder; inputting the extracted each feature into an input layer of a deep artificial neural network, and storing a prediction model generated by synthesizing each feature input to the deep artificial neural network; and predicting the stock price of a stock item included in the KOSPI index by using the stored prediction model. In the step of storing the prediction model, a prediction model can be generated by learning the stacked noise elimination auto-encoder and the deep artificial neural network.

Description

FEATURE LEVEL FUSION SYSTEM AND METHOD FOR STOCK PRICE FORECASTING}

The following description relates to stock price prediction technology, and relates to a method and system for predicting stock price based on feature fusion.

Machine learning technology is widely used to predict stock prices. It is common to use data unique to Korean companies, such as technical indicators, trading data, and accounting data, to predict the stock price of Korean stocks.

For example, Korean Patent Publication No. 10-2014-0120416 relates to a technology for predicting future stock prices through market capitalization analysis, and builds a database by collecting and storing market capitalization information and financial information for each company over a certain period of time. Market capitalization information and financial information by extracting market cap and financial information from the database for a certain period in the past for the stock quotes for which future prediction questions have been received. Judging the similarity between the two, identify financial indicators related to the movement of market capitalization, determine the basic forecasting model for multi-faceted evaluation of stock prices by setting weights and application rules to be assigned to financial information based on the identification results Disclosed is a structure to calculate virtual market cap information to answer a question through a basic predictive model.

However, the Korean stock market is heavily influenced by foreign markets, such as the United States and Europe. Accordingly, in predicting the share price of a Korean company, a technique for increasing prediction accuracy by using data from a foreign stock market is required.

It is possible to provide a forecasting system and method to increase the accuracy of forecasting Korean stock prices by efficiently combining various international stock market data affecting the Korean stock market. Specifically, it is intended to extract important features using deep learning technology and to improve prediction accuracy by using the extracted features for stock price prediction.

The stock price prediction method performed by the prediction system includes: extracting each feature as learning the KOSPI index data and the foreign index data using a multi-layer noise reduction autoencoder; Inputting each of the extracted features into an input layer of a deep artificial neural network, and storing a prediction model generated as each feature input to the deep artificial neural network is synthesized; And predicting a stock price of an item included in the KOSPI index using the stored prediction model, and in the storing of the prediction model, learning the stacked noise removal autoencoder and the deep artificial neural network to obtain a prediction model. Can be created.

In the step of extracting each feature, a pre-processing of scaling the KOSPI index data and the S&P 500 index data to a value between 0 and 1 is performed, and a plurality of the KOSPI index data and the S&P 500 index data subjected to the pre-processing are stacked. The method may include extracting each feature based on unsupervised learning using a noise autoencoder.

In the step of storing the prediction model, error data is generated by applying noise to the KOSPI index data, and a feature is extracted from the hidden layer using an N (N is a natural number) autoencoder composed of an input layer, a hidden layer, and an output layer. , Step of constructing a stacked denoising autoencoder (SDAE) stacking features extracted from a plurality of hidden layers by repeating the process of inputting the features extracted from the hidden layer into the input layer of the N+1 autoencoder. It can contain.

In the storing of the prediction model, each feature, which is output data output through the stacked noise canceling autoencoder, is connected to an input layer of the FNN neural network, and the stacked noise canceling autoencoder and a FNN neural network (SDAE-FNN) ), and training a neural network (SDAE-FNN) composed of the stacked noise canceling autoencoder and FNN based on a backpropagation algorithm.

The step of predicting the stock price of the stock included in the KOSPI index using the stored prediction model includes: of the stock listed on the KOSPI 200 index using the predictive model constructed by learning the multi-layer noise autoencoder and the deep artificial neural network. Including the step of predicting the stock price, the stock price of the stock listed on the KOSPI 200 index may be predicted by simultaneously using the foreign index data and data related to the stock listed on the KOSPI 200 index.

Predicting the stock price of an item included in the KOSPI index using the stored prediction model includes verifying the prediction accuracy of the stock price as the stock price of the item listed on the KOSPI 200 index is predicted. Prediction accuracy can be evaluated by averaging the sum of the measured values of whether the predicted stock price movement predicting the stock price of the listed stock and the actual stock price movement match.

In order to execute the stock price prediction method performed by the prediction system, the computer program stored in the computer-readable storage medium extracts each feature as learning the KOSPI index data and the foreign index data using a stacked noise removal autoencoder. ; Inputting each of the extracted features into an input layer of a deep artificial neural network, and storing a prediction model generated as each feature input to the deep artificial neural network is synthesized; And predicting a stock price of an item included in the KOSPI index using the stored prediction model, and in the storing of the prediction model, learning the stacked noise removal autoencoder and the deep artificial neural network to obtain a prediction model. Can be created.

The prediction system includes a feature extracting unit that extracts each feature as the Kospi index data and the foreign index data are learned using a stacked noise reduction autoencoder; A prediction model storage unit that inputs each of the extracted features into an input layer of a deep artificial neural network, and stores a prediction model generated as each feature input to the deep artificial neural network is synthesized; And a stock price prediction unit for predicting a stock price of an item included in the KOSPI index using the stored prediction model.

The feature extraction unit performs pre-processing for scaling the KOSPI index data and the S&P 500 index data to a value between 0 and 1, and uses a plurality of stacked noise autoencoders for the KOSPI index data and the S&P 500 index data on which the preprocessing was performed. Therefore, each feature can be extracted based on unsupervised learning.

The prediction model storage unit generates noise data by applying noise to the KOSPI index data, extracts features from the hidden layer using an N (N is a natural number) autoencoder composed of an input layer, a hidden layer, and an output layer, and the hidden layer A stacked denoising autoencoder (SDAE) can be constructed by stacking features extracted from a plurality of hidden layers by repeating the process of inputting the features extracted from the input layer of the N+1 autoencoder.

The prediction model storage unit connects each feature, which is output data output through the stacked noise canceling autoencoder, to the input layer of the FNN neural network, and builds a neural network (SDAE-FNN) composed of the stacked noise canceling autoencoder and FNN. And, based on the Backpropagation algorithm, it is possible to train a neural network (SDAE-FNN) composed of the stacked noise canceling autoencoder and FNN.

The stock price prediction unit includes predicting a stock price of an item listed on the KOSPI 200 index using the multi-layer noise autoencoder and the predictive model constructed by learning the deep artificial neural network, and the foreign index data and the KOSPI 200 index. Using the data related to the listed stocks at the same time, the stocks listed on the KOSPI 200 index can be predicted.

The stock price prediction unit includes verifying the prediction accuracy of a stock price listed on the KOSPI 200 index, and predicting stock price and real stock price movement predicting the stock price listed on the KOSPI 200 index. Prediction accuracy can be evaluated by averaging the sum of the measured values of whether or not the match is made.

The prediction system according to an embodiment may provide a future stock price, which is essential for safe asset management, to detect risks early through a portfolio built on high predictive power and obtain a stable return.

The prediction system according to an embodiment may consider various variables affecting the stock price by using a stacked noise reduction autoencoder, derive effects such as overfitting prevention, model complexity relaxation, and computational complexity reduction, and share price prediction accuracy Can increase

1 is a view for explaining the operation of the prediction system according to an embodiment.
2 is a block diagram illustrating the configuration of a prediction system according to an embodiment.
3 is a flowchart illustrating a stock price prediction method of a prediction system according to an embodiment.
4 is a diagram for explaining the construction of a stacked noise cancellation autoencoder in a prediction system according to an embodiment.
5 is a diagram for explaining a method of extracting features through a stacked noise removal autoencoder in a prediction system according to an embodiment.
6 is a graph for explaining prediction accuracy in a prediction system according to an embodiment.

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

1 is a diagram for describing an operation of a prediction system according to an embodiment.

The prediction system 100 is for predicting a stock price based on feature convergence, and can increase accuracy in predicting a stock price of a Korean company by using data of a domestic stock market as well as data of a foreign stock market. The prediction system 100 may combine stock market data opened at different time periods with domestic index data to predict the stock price of the stock listed on the KOSPI index.

The prediction system 100 may input data to build a stock price prediction model. For example, the input data may be trading (transaction) data related to the starting price, closing price, low price, high price and trading volume of the KOSPI index data (domestic index data) 101 and the foreign index data 111. Some of each of these data can be used to train, and can be used to evaluate the remaining predictive model of each data. The foreign index data may refer to foreign index data such as the United States, Japan, and China. In an embodiment, at least one foreign index data may be used as the foreign index data, and the S&P 500 index data will be described as an example of the foreign index data. US stocks affect the world's stocks. The prediction system 100 may predict what will be included in the domestic KOSPI index using S&P 500 index data related to the U.S. stock market, which closes at around 7 am based on Korea.

The prediction system 100 may perform pre-processing (120, 121) of scaling input data to values between 0 and 1 in order to convert it into data suitable for deep learning. At this time, pre-processing is not limited to scaling, and various pre-processing processes may be included in addition to scaling. For example, the prediction system may preprocess each of the KOSPI index data and the foreign index data.

The prediction system 100 may extract each feature by learning each of the pre-processed KOSPI index data and the pre-processed foreign index data through the stacked noise removal autoencoders 130 and 131. In this case, as the stacked noise canceling autoencoder, a plurality of stacked noise canceling autoencoders (for example, two stacked noise canceling autoencoders) may be used. For example, the prediction system 100 may extract high-dimensional characteristics of the data by non-supervising learning the pre-processed KOSPI index data and the pre-processed foreign index data using each layered noise autoencoder.

The prediction system 100 may synthesize the features of the two indices (KOSPI index data and foreign index data) by inputting each extracted feature into an input deep-neural network of the deep neural network 141. A deep feedforward neural network (FNN) may be used as the deep neural network 141. The prediction model 160 may be constructed based on the synthesized features. At this time, a learning model may be generated as the plurality of stacked noise canceling auto-encoders 130 and 131 and the deep artificial neural network 141 are entirely supervised and trained. In other words, the values can be fine-tuned as the stacked noise canceling auto-encoders 130 and 131 and the deep artificial neural network 141 are learned as a whole.

As an example, the prediction model 160 may be constructed as learning is performed through the multilayer noise canceling auto-encoders 130 and 131 and the deep artificial neural network 141. At this time, the S&P index data 111, which is the foreign index data obtained by closing the chapter of the day, may be input to be reflected in the prediction model 160. The stock 150 data listed on the KOSPI 200 index in the predictive model 160 constructed as described above may be input to the predictive model 160.

As another example, a predictive model benchmarked with respect to S&P index data may exist. The predicted model 160 that is appropriately modified to the KOSPI index using the learning model generated by performing the training through the stacked noise removing auto-encoders 130 and 131 and the deep artificial neural network 141 is performed. Can build. The stock 150 data listed on the KOSPI 200 index may be input to the prediction model 160.

As another example, the prediction model 160 may be constructed as learning is performed through the stacked noise canceling auto-encoders 130 and 131 and the deep artificial neural network 141. The prediction system 100 may perform prediction and analysis by simultaneously inputting S&P index data 111 and stock 150 listed on the KOSPI 200 index into the prediction model 160.

The prediction system may predict the stock price of the stock listed on the KOSPI 200 index using the prediction model 160. The prediction system may perform analysis on the stock 150 listed on the KOSPI 200 index as data of the stock 150 listed on the KOSPI 200 index is input to the prediction model 160. For example, the prediction system may determine whether the stock price will rise or fall for each of the stocks 150 listed on the KOSPI 200 index. For example, the prediction system may predict the stock price of an item listed on the KOSPI 200 index of the day, the next day, or a preset date using the prediction model 160. The prediction system may update data predicting the stock price of the stocks by updating the stock price of the stocks listed on the KOSPI 200 index every predetermined time. Alternatively, the prediction system may update the prediction for the stock price of the stock listed on the KOSPI 200 index as new data is input. The prediction system can verify the prediction accuracy 170 for the predicted prediction data through the prediction model 160.

2 is a block diagram illustrating a configuration of a prediction system according to an embodiment, and FIG. 3 is a flowchart illustrating a stock price prediction method of a prediction system according to an embodiment.

The processor 200 of the prediction system 100 may include a feature extraction unit 210, a prediction model storage unit 220, and a stock price prediction unit 230. The components of the processor 200 may be expressions of different functions performed by the processor 200 according to a control command provided by the program code stored in the prediction system. The processor 200 and components of the processor 200 may control the prediction system to perform steps 310 to 330 included in the stock price prediction method of FIG. 3. At this time, the processor 200 and components of the processor 200 may be implemented to execute instructions according to code of an operating system included in a memory and code of at least one program.

The processor 200 may load program code stored in a file of a program for a stock price prediction method into a memory. For example, when a program is executed in the prediction system, the processor may control the prediction system to load program code from a file of the program into a memory under the control of the operating system. At this time, the processor 200 and the feature extraction unit 210, the prediction model storage unit 220, and the stock price prediction unit 230 included in the processor 200 each command an instruction of a corresponding portion of the program code loaded in the memory. It may be different functional representations of the processor 200 to execute to execute the subsequent steps 310 to 330.

In step 310, the feature extraction unit 210 may extract each feature by learning the KOSPI index and the foreign index using a stacked noise reduction autoencoder. The feature extraction unit 2100 performs pre-processing to scale the KOSPI index data and the S&P 500 index data to values between 0 and 1, and uses a plurality of stacked noise autoencoders to perform the pre-processed KOSPI index data and S&P 500 index data. Therefore, each feature can be extracted based on unsupervised learning.

In step 320, the prediction model storage unit 220 may input each extracted feature into the input layer of the deep neural network, and store the generated prediction model as each feature input to the deep artificial neural network is synthesized. . The predictive model storage unit 220 may generate a predictive model by learning the multilayer noise canceling autoencoder and the deep artificial neural network. The prediction model storage unit 220 generates noise data by applying noise to the KOSPI index data, extracts features from the hidden layer using an N (N is a natural number) autoencoder composed of the input layer, the hidden layer, and the output layer, and the hidden layer A stacked denoising autoencoder (SDAE) can be constructed by stacking features extracted from a plurality of hidden layers by repeating the process of inputting the features extracted from the input layer of the N+1 autoencoder.

The predictive model storage unit 220 connects each feature, which is output data output through the layered noise canceling autoencoder, to the input layer of the FNN neural network, and constructs a neural network (SDAE-FNN) composed of the layered noise canceling autoencoder and FNN. And, based on the Backpropagation algorithm, it is possible to train a neural network (SDAE-FNN) composed of a stacked noise canceling autoencoder and FNN.

In step 330, the stock price prediction unit 230 may predict the stock price of the item included in the KOSPI index using the stored prediction model. The stock price prediction unit 230 may predict the stock price of an item listed on the KOSPI 200 index using a stacked noise autoencoder and a prediction model constructed by learning the deep artificial neural network. The stock price prediction unit 230 may verify the prediction accuracy of the stock price as the stock price of the stock listed on the KOSPI 200 index is predicted. The stock price predicting unit 230 may evaluate the prediction accuracy by averaging the sum of the measured values of whether the predicted stock price movement predicted the stock price of the stock listed on the KOSPI 200 index coincides with the realized stock price movement.

4 is a diagram for explaining the construction of a stacked noise cancellation autoencoder in a prediction system according to an embodiment.

The process of constructing a multilayer noise canceling autoencoder will be described. By applying a certain noise to the KOSPI index data, it is possible to generate crashed data.

In general, an autoencoder means an artificial neural network capable of learning an efficient code (for example, encoding) of input data using unlabeled training data without any supervision. The autoencoder may be composed of an encoder (cognitive network) that converts inputs to internal representations and a decoder (generation network) that converts internal representations to outputs. The autoencoder may have a plurality of hidden layers, and this case is referred to as a stacked autoencoder or deep autoencoder. The autoencoder may be composed of an input layer, a hidden layer, and an output layer. The auto-encoder is optimized by minimizing the error between the input data of the input layer in the auto-encoder and the output data of the output layer in the auto-encoder, and in this process, the hidden layer extracts the level characteristics above a predetermined reference (high dimensional). do.

A stacked noise removal autoencoder may be configured by stacking a plurality of autoencoder hidden layers for a unit autoencoder composed of an input layer, a hidden layer, and an output layer. The stacked noise canceling autoencoder generates noise data by applying noise to the KOSPI index data, extracts features from the hidden layer using the Nth N (N is a natural number) autoencoder composed of the input layer, the hidden layer, and the output layer. , A stacked denoising autoencoder (SDAE) can be constructed by stacking features extracted from a plurality of hidden layers by repeating the process of inputting features extracted from the hidden layer into the input layer of the N+1 autoencoder. In other words, the feature extracted from the hidden layer using the unit autoencoder can be used as the input layer of the next auto encoder.

The prediction system may acquire characteristics of a predetermined level or higher (high-level) by stacking compressed characteristics of the KOSPI index data extracted from the hidden layer using an autoencoder. Likewise, by using the autoencoder to accumulate the compressed features of the S&P 500 index data extracted from the hidden layer, it is possible to obtain features above a predetermined criterion.

5 is a diagram for explaining a method of extracting features through a stacked noise removal autoencoder in a prediction system according to an embodiment.

The prediction system can build a hybrid model (SDAE-FNN) model by connecting a stacked noise canceling autoencoder to the input layer of a deep feedforward neural network (FNN). In a feedforward neural network, information flow in a multi-layer perceptron starts from the input layer and goes through the hidden layer to the output layer. The prediction system can train a hybrid model (SDAE-FNN) model using a backpropagation algorithm. Backpropagation is a representative supervised learning algorithm used to train a feedforward neural network (FNN) with multiple hidden layers with labeled learning data. In an embodiment, the FNN is an example of a deep neural network for connecting a multilayer noise canceling autoencoder, but is not limited to the FNN.

Referring to FIG. 5, it is shown that the high level feature of the multilayer noise canceling autoencoder is extracted. Specifically, each stacked noise removal autoencoder for KOSPI index data and foreign index data (for example, S&P 500) is shown. The output data output through each layered noise canceling autoencoder can be connected to the input layer of the FNN neural network (fusion layer shown in FIG. 5). The entire neural network (hybrid model) composed of a stacked noise canceling encoder and FNN can tune hyperparameters based on the backpropagation algorithm.

6 is a graph for explaining prediction accuracy in a prediction system according to an embodiment.

In a prediction system, prediction accuracy can be evaluated using a hit ratio. It can be defined as 1 when the predicted share price movement including rising or falling coincides with the realized stock price movement and 0 when the predicted stock price movement does not coincide with the realized stock price movement. The prediction system can verify the prediction accuracy through the hit ratio by averaging the sum of the measured values of whether the predicted stock price movement predicted the stock price and the realized stock price movement match.

As an example, the prediction system can not only predict the stock price for each of the stocks listed on the KOSPI 200 index, but also predict the stock price of the industry including the stock. The prediction system can classify items listed on the KOSPI 200 index based on predetermined criteria. The prediction system can classify the stocks based on the same or similar industries for the stocks listed on the KOSPI 200 index. For example, the prediction system may classify items based on the industry type, such as semiconductor stock, cosmetic stock, and bio stock. The prediction system may classify the predicted data predicted through the predictive model for each item listed on the KOSPI 200 index, and predict the stock price by industry by verifying the prediction accuracy of the classified predicted data.

Referring to FIG. 6, the hit ratio distribution of the next day's stock price prediction accuracy for a company belonging to KOSPI 200 is shown. The average value to be evaluated is about 0.73, and the predictive model can evaluate that the next day's share price rise/fall can be predicted with 73% accuracy.

The device described above may be implemented with hardware components, software components, and/or combinations of hardware components and software components. For example, the devices and components described in the embodiments include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors (micro signal processors), microcomputers, field programmable gate arrays (FPGAs). , A programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose computers or special purpose computers. The processing device may run an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process 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 embodied in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored in 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 on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded in the medium may be specially designed and configured for the embodiments or may be known and usable by 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, DVDs, and magnetic media such as floptical disks. -Hardware devices specifically 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 high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and/or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (13)

In the stock price prediction method performed by the prediction system,
Extracting each feature as learning the KOSPI index data and the foreign index data using a stacked noise reduction autoencoder;
Inputting each of the extracted features into an input layer of a deep neural network, and storing a prediction model generated as each feature input to the deep artificial neural network is synthesized; And
Predicting the stock price of the stock included in the KOSPI index using the stored prediction model
Including,
In the step of storing the predictive model, the predictive model is generated by learning the multilayer noise canceling autoencoder and the deep artificial neural network.
A stock price prediction method that includes things. According to claim 1,
Extracting each of the features,
Based on unsupervised learning by performing pre-processing to scale KOSPI index data and S&P 500 index data to values between 0 and 1, and using the multi-layered noise autoencoder to perform the pre-processed KOSPI index data and S&P index data. Steps to extract each feature
Stock price prediction method comprising a. According to claim 1,
The step of storing the prediction model,
The noise is generated by applying noise to the KOSPI index data, and a feature is extracted from the hidden layer using an N (N is a natural number) autoencoder composed of an input layer, a hidden layer, and an output layer, and a feature extracted from the hidden layer is removed. Constructing a stacked denoising autoencoder (SDAE) that stacks features extracted from a plurality of hidden layers by repeating the input process to the input layer of the N+1 autoencoder.
Stock price prediction method comprising a. According to claim 1,
The step of storing the prediction model is
Each feature, which is output data output through the multilayer noise canceling autoencoder, is connected to the input layer of the FNN neural network, a neural network (SDAE-FNN) composed of the multilayer noise canceling autoencoder and FNN is built, and based on a backpropagation algorithm. Learning the neural network (SDAE-FNN) composed of the stacked noise canceling auto-encoder and FNN.
Stock price prediction method comprising a. According to claim 1,
Predicting the stock price of the stock included in the KOSPI index using the stored prediction model,
Predicting the stock price of the stock listed on the KOSPI 200 index using the stacked noise autoencoder and the predictive model constructed by learning the deep artificial neural network.
Including,
The stock price of an item listed on the KOSPI 200 index is predicted by simultaneously using the foreign index data and data related to the item listed on the KOSPI 200 index.
How to predict stock prices. The method of claim 5,
Predicting the stock price of the stock included in the KOSPI index using the stored prediction model,
Verifying the prediction accuracy of the stock price as the stock price of the stock listed on the KOSPI 200 index is predicted
Including,
Predictive accuracy is evaluated by averaging the sum of the measured values of whether the predicted stock price movement and the realized stock price movement are predicted by predicting the stock price of the stock listed on the KOSPI 200 Index.
How to predict stock prices. A computer program stored in a computer-readable storage medium for executing the stock price prediction method performed by the prediction system,
Extracting each feature as learning the KOSPI index data and the foreign index data using a stacked noise reduction autoencoder;
Inputting each of the extracted features into an input layer of a deep neural network, and storing a prediction model generated as each feature input to the deep artificial neural network is synthesized; And
Predicting stock prices of stocks included in the KOSPI index using the stored prediction model
Including,
In the step of storing the predictive model, the predictive model is generated by learning the multilayer noise canceling autoencoder and the deep artificial neural network.
A computer program stored in a computer readable storage medium comprising a. In the prediction system,
A feature extraction unit for extracting each feature as learning the Kospi index data and the foreign index data using a stacked noise reduction autoencoder;
A prediction model storage unit that inputs each of the extracted features into an input layer of a deep neural network, and stores a prediction model generated as each feature input to the deep artificial neural network is synthesized; And
A stock price prediction unit that predicts the stock price of an item included in the KOSPI index using the stored prediction model
Including,
In the prediction model unit, learning the stacked noise removal autoencoder and the deep artificial neural network to generate a prediction model
Prediction system. The method of claim 8,
The feature extraction unit,
Based on unsupervised learning by performing pre-processing to scale the KOSPI index data and S&P 500 index data to values between 0 and 1, and using the multi-layered noise autoencoder to perform the pre-processed KOSPI index data and S&P index data. To extract each feature
Prediction system characterized in that. The method of claim 8,
The prediction model storage unit,
The noise is generated by applying noise to the KOSPI index data, and a feature is extracted from the hidden layer using an N (N is a natural number) autoencoder composed of an input layer, a hidden layer, and an output layer, and a feature extracted from the hidden layer is removed. By constructing a stacked denoising autoencoder (SDAE) that stacks features extracted from multiple hidden layers by repeating the input process to the input layer of the N+1 autoencoder.
Prediction system characterized in that. The method of claim 8,
The prediction model storage unit,
Each feature, which is output data output through the multilayer noise canceling autoencoder, is connected to the input layer of the FNN neural network, a neural network (SDAE-FNN) composed of the multilayer noise canceling autoencoder and FNN is built, and based on a backpropagation algorithm. To train the neural network (SDAE-FNN) composed of the stacked noise canceling autoencoder and FNN.
Prediction system characterized in that. The method of claim 8,
The stock price prediction unit,
Predicting the stock price of the stock listed on the KOSPI 200 index using the stacked noise autoencoder and the predictive model constructed by learning the deep artificial neural network,
The stock price of an item listed on the KOSPI 200 index is predicted by simultaneously using the foreign index data and data related to the item listed on the KOSPI 200 index.
Prediction system characterized in that. The method of claim 12,
The stock price prediction unit,
Verifying the prediction accuracy of the stock price as the stock price of the stock listed on the KOSPI 200 index is predicted,
Prediction accuracy is evaluated by averaging the sum of the measured values of whether the predicted stock price movement predicted the stock price of the stock listed on the KOSPI 200 index coincides with the realized stock price movement
Prediction system characterized in that.

Images