Neural Network Approach for Stock Prediction using Historical Data

Transcription

1 Neural Network Approach for Stock Prediction using Historical Data Yuvraj Wadghule SND COE & RC,Yeola Prof. I.R. Shaikh SND COE & RC,Yeola ABSTRACT In today s era the count of investor is increasing dayby day. For identifying the market risk and for the growth of profit the market stock prediction is important factor. Peoples are using traditional theorems for predicting market behavior. But due to high risk in market and the world is advancing, accurate result matter for the profit. Different methods are used for predicting the market share price such as support vector machine, Regression, Sentiments from different social websites like Twitter and Facebook which have some limitation. In this paper we are specifying the artificial neural network with back propagation and statistical regression.in this the historical data is used to train the neurons. The normalized value obtained to feed to neuron for training them is quite efficient. By using this method it will help the investor, trader to invest in market and exit from the market by using the historical data of the stock. This method will have good accuracy for prediction of sharemarket. Keywords: Stock Prediction, ANN, Regression, Support vector machine, Back Propagation. INTRODUCTION Industry permits businesses to be traded publicly by selling shares of ownership of the company in a public market. Stock market can affect the social mood to a great extent. An economy where the stock market is on the hike is considered to be an upcoming economy. Rising prices of shares is generally linked with increased business investment and vice versa. The prediction helps in decision making process whether to buy or sell a share. The vital idea to successful stock market prediction is to achieve best results and minimize the inaccurate forecast the stock price. It is veryimportant to predict the stock as it may lead to high profit or huge loses that the buyer/seller has to bear. Forecasting stock indices is very difficult because of the market volatility that needs accurate forecast model. The stock market indices are highly fluctuating that s fall the stock price or raising the stock price. Fluctuations are affecting the investor s belief. Determining more effective ways of stock market index prediction is important for stock market investor in order to make more informed and accurate investment decisions. Prediction of stock market is not a simple task, because of the variation in the behavior of the stock price in short time period. Number of techniques are being used for stock market prediction. In last twenty years artificial neural networks (NNs) have become one of these techniques. As the behavior of stock price variation is highly complex, so it is not possibleto model a pure mathematical function for it. Forecasting stock indices is very difficult because of the market volatility that needs accurate forecast model. The stock market indices are highly fluctuating that fall the stock price or raising thestock price. Fluctuations are affecting the investor s belief.determining more effective ways of stock market index prediction is important for stock market investor in order to make more informed and accurate investment decisions. Huge amount of data set is required for explaining a specific stock value. As stock market domain is unique and unpredictable domain, there are different methods like statistical analysis, fundamental analysis and technical analysis with each having different results. This methods cannot provide the accurate result in efficient manner. In this 636 Yuvraj Wadghule, Prof. I.R. Shaikh

2 paper we are combining two techniques first one is regression for predicting values of any entity and second one is back propagation using neural network. REVIEW OF LITERATURE R.K. and Pawar D.D. in [1] discussed the stock prediction methods to find out which is more effective and accurate method so that a buy or sell signal can be generated for given stocks. Predicting stock index with traditional time series analysis has proven to be difficult but artificial neural network may be suitable. Neural network has the ability to extract useful information from large set of data. The paper also discusses a literature review on application of artificial neural network in stock market Index prediction. Halbert white in [2] reported some results of an on-going project using neural network modeling and learning techniques to search for and decode nonlinear regularities in asset price movements. Author, focus on case of IBM common stock daily returns. Tiffany Hui-Kuang and Kun-Huang Huarng in [3] used neural network for handling nonlinear relationship and given a new fuzzy time series model to improve forecasting. The fuzzy relationship is used to forecast the Taiwan stock index. In the neural network fuzzy time series model is used for forecasting. The drawback of taking all the degree of membership for training and forecasting may affect the performance of the neural network. B. Chauhan, U. Bidave, A. Gangathade, and S. Kale in [4] discussed the machine learning approach for data mining to evaluate the predictive relationship of numerous financial and economic variables. A crossvalidation technique was used to improve the generalization ability of several models. The author decides to deploy the forecast the stock dividends, transaction costs and individual-tax brackets to replicate the realistic investment practices. Sam Mahfoud and Ganesh Mani [5] propose financial forecasting using genetic algorithms a new system that utilizes genetic algorithms (GAs) to predict the future performances of individual stocks. More generally, the system extends GAs from their traditional domain of optimization to inductive machine learning or classification. The overall learning system incorporates a GA, a niching method and several other components. For stock market prediction K. Senthamarai Kannan, P. Sailapathi Sekar, M.Mohamed Sathik and P. Arumugam [6] used diffrent data mining techniques. In the data mining the main thing is historic data that holds the essential memory for predicting the future direction. From the historic stock market data investors discovers the hidden pattern of the data that have predictive capability in the investment decisions. Using data mining technology. In financial time series prediction the prediction of stock market is regarded as challenging task. We can estimate future stock price increase or decrease by the data analysis. Data analysis is also one way of prediction. Five methods of analyzing stocks were combined to predict. Typical Price (TP), Relative Strength Index (RSI), Bollinger Bands, Moving Average (MA) and CMI are the proposed five methods. Combing these methods would be useful for predicting day s closing price would increase or decrease. Aditya Gupta and Bhuvan Dhingra[7] used Hidden Markov Model(HMMs) for the predicting the stock market. By using historical stock prices they present he Maximum a Posteriori HMM approach for forecasting stock values for the next day. For training the continuous HMM they consider the intraday high and low values of the stock and fractional change in stock values. Over all the possible stock values for the next day this HMM is used to make a maximum posteriori decisions. By using some of the existing methods like HMMs and Artificial Neural Networks using Mean Absolute Percentage error (MAPE). They test their approach on several stocks, and compare the performance. Finally they present an HMM based Maximum a Posteriori (MAP) estimator for stock predictions. The model uses a latency of days to predict the stock value for the (d + 1) st day. Using a previously trained continuous HMM MAP decision is made over all the possible values of stock. Nair, Mohandas and Sakthivel [8] proposes a decision tree rough set hybrid system for stock market prediction presents the design and performance evaluation of a hybrid decision tree- rough set based system for predicting the next day s trend in the Bombay Stock Exchange (BSESENSEX). This system outperforms both the neural network based system and the nave bayes based trend prediction system. 637 Yuvraj Wadghule, Prof. I.R. Shaikh

3 D. V. Setty, T. M. Rangaswamy, and K. N. Subramanya in [9] used the moving average (MA) method to uncover the patterns, relationship and to extract values of variables from the database to predict the future values of other variables through the use of time series data. The advantage of the MA method is a device for reducing fluctuations and obtaining trends with a fair degree of accuracy. M. Suresh Babu, N. Geethanjali and B. Sathyanarayana in [10] used the data mining techniques used to uncover the hidden pattern, predict future trends. Apart from segment size the ant to sub-time-series size affects the system performance. Y.L. Hsieh, Don-Lin Yang and Jungpin Wu in [11] used data mining methods for association rule and sequential pattern mining. Association rule can be used to analyze the customer consumption behaviors and find the patterns of buying habits in the retailer business. The sequential pattern was used to help web viewers match their needs quickly. The author said that the usage of the association rule and sequential pattern mining methods are extend of causal relationship chain and improved the accuracy level. PROBLEM STATEMENT Although there are various techniques implemented for the prediction of stock market. On the basis of existing technique used for the future prediction of stock market a new technique for the prediction is proposed which provides close prediction of stock market. In this paper I have proposed the two methods which are combine and used for stock market prediction. As the study shows that only neural network which learn from historical data will provide better result but by applying regression method will provide good and more accurate result. SYSTEM ARCHITECTURE To provide the user the best result for predicting the stock prices and gaining the profit I have proposed the working model that will be useful for prediction using historical data. The linear regression method and Back propagation Neural network are applied are final result can be prediction as shown in below figure: 638 Yuvraj Wadghule, Prof. I.R. Shaikh Fig. 1. Architecture of Proposed System

4 Architecture of Proposed System consist of four basic modules and the working of each module is explained in detail as below: A. DATA COLLECTION AND PRE-PROCESSING In data collection we are collecting the input data. The input data may be a csv file that may consist of stock market record of recent few year which is our historical data.after data collection we are pre-processing that input data file. Preprocessing is done on input csv file to eliminate following attribute. Handle null and empty record by assigning new valuescomputed through linear regression. Invalid data records are replaced by valid and approximatedata records. The data records of non-interest are removed from thecsv file. After preprocessing the Normalization on data is done. Normalization is statistical method to smoothen the range of values in the numeric data records. Normalization with standard deviationtransforms variable data range into -3 to 3.Normalization helps in machine learning where a small drift into input dataset affects the results. B. LINEAR REGRESSION Regression predicts a numerical value. Regression performs operations on a dataset where the target values have been defined already. And the result can be extended by adding new information. The relations which regression establishes between predictor and target values can make a pattern. Thispattern can be used on other datasets which their target values are not known. Therefore the data needed for regression are 2 part, first section for defining model and the other for testing model. In this section we choose linear regression for our analysis. First, we divide the data into two parts of training and testing. Then we use the training section for starting analysis and defining the model. Scatter plot of 80% out of data has been shown in (figure 1) with taking this into consideration that the (Average) parameter is the mean of the prices of Open, Low, High and close. Scatter plot has been shown with just the Average parameter in order to be simpler. Fig. 2. LINEAR REGRESSION LINE C. ANN NETWORK USING BACK PROPAGATION Artificial Neural Networks (ANN) actually ha s the potential to tell apart unknown and hidden patterns in information which may be terribly effective for share market prediction. BP network is that the backpropagation network. It s a multilayer forward network, learning by minimum mean sq. error. It may be employed in the sphere of language integration, identification and adaptation management, etc. BP network is 639 Yuvraj Wadghule, Prof. I.R. Shaikh

5 semi supervised learning. Initial of all, artificial neural network has to learn an exact learning criteria, so it will work. If the result yielded by network is wrong, then the network ought to scale back the chance of creating identical mistake next time through learning. This paper uses data processingtechnique to check historical information concerning share market in order that it will predict the desired values a lot of accurately. D. PREDICTION This module refers to extract outcomes from both (hybrid) approaches of linear regression and ANN and finally predict the most accurate value. The accuracy is measured in training phase and then used to control the threshold of prediction for every sample in testing phase. ALGORITHM USED FOR IMPLEMENTATION Neural network using Back Propagation Accept input sample Perform its weighted summation. Apply it to input layer neurons. Process all inputs at each neuron by transfer function to get individual. Hidden layer and repeat 1,2,3,4 steps pass it as an input to all neurons of for hidden layer neurons. Pass output of hidden layer neurons to all output layers and repeat 1,2,3,4 steps to get final output. Display the final output. MATHEMATICAL MODEL Error calculation: Calculating Root Mean Square, Let RMS is denoted as Root Mean Square, E is denoted as Error of difference between actual value and predicted value GE means Global Error Updating, error value, Where, delta=expected value-actual value. Activation function: Sigmoid Result=1/Tan hyperbolic RESULT & DISCUSSION Expected result of stock analysis using regression method is shown below fig 3. The result by this method is not so accurate. Then by applying ANN by back propagation the result is improved in fig 4.But by combining both methods we will get the accurate prediction shown in fig. 5. Table 1. RESULT USING REGRESSION ALGORITHM Date Open Price Close Price Difference 26/12/ /12/ /2/ /12/ /1/ Yuvraj Wadghule, Prof. I.R. Shaikh

6 Fig:3 TABLE II: RESULT USING REGRESSION ALGORITHM Date Open Price Close Price Difference Dec. Shifting 26/12/ /12/ /2/ /12/ /1/ The proposed system will give us the following result. Result of stock market analysis by linear regression. Result by using Artificial Neural Network. Accurate result after using both ANN and linear regression. Fig Yuvraj Wadghule, Prof. I.R. Shaikh

7 TABLE IV:RESULT USING THE NEW PROPOSED APPROACH THAT IS REGRESSION Date Open Price Close Price Difference Dec. Shifting 26/12/ /12/ /2/ /12/ /1/ Fig 5 CONCLUSION In this paper, we tried to sum up the application of Artificial Neural Networks (ANN) for predicting stock market with linear regression method Using both method provides a good and accurate result. Back propagation algorithm is the best algorithm to be used in Feed forward neural network because it reduces an error between the actual output and desired output in a gradient descent manner. It will help private investors and stakeholders to invest in stock market. Calculation through Regrogation gave us optimized normalized value. This approach is easy to understand and bears less complexity. The future scope includes integration of Regrogation with cloud. REFERENCES [1] Dase R.K. and Pawar D.D., Application of Artificial Neural Network for stock market predictions: A review of literatureinternational Journal of Machine Intelligence, ISSN: , Volume 2, Issue 2, 2010, pp [2] albert White, Economic prediction using neural networks: the case of IBM daily stock returns Department of Economics University of California, San Diego. [3] Tiffany Hui-Kuang yu and Kun-Huang Huarng, A Neural network-based fuzzy time series model to improve forecasting, Elsevier, 2010, pp: [4] B. Chauhan, U. Bidave, A. Gangathade, and S. Kale, Stock Market Prediction Using Artificial Neural Networks, International Journal of Computer Science and Information Technologies, 2014, pp [5] Sam Mahfoud and Ganesh Mani, Financial Forecasting using Genetic Algorithms, Applied Artificial Intelligence, 10: pp , Yuvraj Wadghule, Prof. I.R. Shaikh

8 [6] K. Senthamarai Kannan, P. Sailapathi Sekar, M. Mohamed Sathik and P. Arumugam,Financial Stock Market Forecast using Data Mining Techniques, Proceedings of the International Multi Conference of Engineer and Computer Scientists 2010 Vol 1, March 17-19,2010,Hong Kong [7] Aditya Gupta, Non Student Member, IEEE and Bhuwan Dhingra, Non Student Member, IEEE, Stock Market Predictions Using Hidden Markov Models [8] Nair, Binoy B., V. P. Mohandas, and N. R. Sakthivel. A decision treerough set hybrid system for stock market trend prediction. International Journal of Computer Applications 6.9 (2010): 16. [9] D. V. Setty, T. M. Rangaswamy, and K. N. Subramanya, A review on Data Mining Applications to the Performance of Stock Marketing, International Journal of Computer Applications, vol. 1, no. 3,pp , Feb [10] M. Suresh babu, N.Geethanjali and B. Sathyanarayana, Forecasting of Indian Stock Market Index Using Data Mining & Artificial Neural Nework, International journal of advance engineering & application, [11] Y.L.Hsieh, Don-Lin Yang and Jungpin Wu, Using Data Mining to study Upstream and Downstream causal relationship in stock Market 643 Yuvraj Wadghule, Prof. I.R. Shaikh