Daily Stock Price Direction Prediction using Random Multi-Layer Perceptron

Abstract

The stock market has always been a quick income source but involved great risks for its high uncertainty. Stock analysts use various fundamental techniques to predict its nature but the results haven't always been profitable. It is mandatory to have a secure prediction method to gain maximum benefits. In this era of automation, machine learning in data science is a valuable tool to predict the nature of the stock market conditions. The literature provides a variety of machine learning techniques such as SVM, AdaBoost, Regression, etc. This study proposes a novel technique called Random MultiLayer Perceptron (RMLP) Classifier which divides the dataset into subsets and applies MLP on them individually. It predicts whether the closing price of the stocks of a particular firm will increase or decrease on the next day by considering the historical data of the firm's stocks as input. This technique gives an accuracy of about 78% which is greater than normal multilayer perceptron in predicting the direction of the stock prices. The proposed method of RMLP is also compared with other existing methods of predicting the direction of the stock prices and promising results are obtained in favor of the proposed method.