Abstract

Several research initiatives have been taken to predict stock market returns using historical data. Investors can find plenty of algorithms that detect the exact closing price of any stock but will not tell the direction of the closing price. During this proposed work, twenty-two years’ price of the stock’s daily close price is being utilized for direction detection. The objective of this paper is to get the right stock, perform exploratory data analysis for data preparation and then build the right models by using multiple modelling techniques to predict whether the price will move up or move down. Closing prices are being utilized as six different feature variables for building the classification model. The difference between the seventh and eighth day closing price is determined. The 0.7%, 1%, and 1.5% differences are different classes of direction to determine either positive, negative or no change. A similar process is again repeated for the feature variable increased to ten days and fourteen days respectively. Then momentum, trend, volatility, and volume indicators are utilized as feature variables and different classification models are built to determine upward direction detection. Random forest model-ling has given the highest efficiency in direction detection. Logistic regression modelling done for percentage change in close price as 0.5% has given the high-est efficiency for volume and momentum indicators whereas the extreme gradient boost classifier provided the best prediction performance for trend and volatility indicators. Therefore, various classification modelling techniques had been re-markably useful in direction detection for the stock under consideration

Keywords: Direction Detection, Stock Market, Technical Indicators, Classification Models, HDFC, KOTAK, SBI.

Conference Published in: 4th international conference on recent trends in communication and intelligent systems (ICRTCI 2023)