Stock price prediction using feature engineering and machine learning techniques

Abstract

The correct prediction of stock prices is a challenging task, as stock prices are affected by a large number of parameters. Moreover, many of these parameters, such as investor sentiment or future market potential, cannot be measured and quantified directly, while having a substantial impact on individual stocks and the stock market as a whole. In this project, I analyzed the changes in the stock price to predict the stock's direction in the future. That is done by extracting multiple descriptors from past data and using them to predict the price change of the stock up to 100 days in the future. Experimental results are collected using 10 stocks and Random Forest, SVM, and KNN classifiers and compared against a baseline ZeroR prediction. The project's goal is to assist the stock traders by providing data-driven insights about the predicted time and direction of changes in the stock price.