OWL REGRESSION BASED TECHNICAL INDICATOR AND INCREMENTAL DECISION DEEP CONVOLUTION FOR STOCK MARKET PREDICTION

Abstract

Stock market analysis is immensely paramount for investors in view of the fact that calculating the future course of action and grasping the fluctuating features of stock prices will reduce the risk of capital investment for profit. In Stock Market Prediction, the objective is to predict future financial stock value of a company. The current trend in stock market prediction is the utilization of Deep Learning (DL) algorithm that makes predictions on the basis of the values of current stock market indices by training on their previous values. By employing DL algorithms prediction of stock market can make prediction easier. In this work a nature inspire optimization and deep convolution learning method called, Owl Search Optimized Deming Regression and Incremental Decision Tree (OSODR-IDT) for stock market prediction is proposed. The OSODR-IDT based deep convolution learning method for stock market prediction is split into one input layer, two hidden layers and finally one output layer. The data collected from stock data collected from the internet, Stock Market Data - Nifty 100 Stocks (1 min) and 2 indices (Nifty 50 and Nifty Bank indices) is provided as input in the input layer. In the first hidden layer or the pooling layer minimizes the spatial dimensions of feature maps using Owl Search Optimized Deming Regression with the intent of obtaining optimal technical indicators to determine the best parameters (i.e. suitable day for buy and sell). With the optimal identification of technical indicators are transmitted to second hidden layer. In the second hidden layer an Incremental Decision Tree-based trading model is performed in the second hidden fully connected layer to generate the trading results. A comprehensive comparative results analysis stated the promising performance of the OSODR-IDT method for stock market prediction over the recent methods in terms of sensitivity, specificity, accuracy, training time, classification error and precision. Simulations performed to validate the proposed method in Python language were found to be improved in terms of precision by 11%..