Predicting the daily closing price of selected shares on the Dhaka Stock Exchange using machine learning techniques

One of the most challenging topics in financial market analysis is predicting stock prices. Factors like supply and demand in the market, market sentiment, and investor's expectations, economic and political shocks can affect stock prices. All these factors make stock prices volatile and chaotic. Several machine learning models have been developed to make more precise and accurate predictions. Support vector regression (SVR) and K-nearest neighbor (KNN) regression are the most popular machine learning techniques used for stock price prediction. Our study follows the hypothesis that the SVR algorithm is a more precise way of predicting the Dhaka Stock Exchange (DSE) than the KNN regression. This research has made predictions and compared prediction errors between SVR and KNN. The analysis has been conducted on recent years’ data of some selected shares listed on the DSE. The performance of the models is measured in terms of their root mean squared error (RMSE), R-squared ( $$R^{2}$$ R 2 ), adjusted R-squared score values. We have optimized our model performance by tuning different combinations of hyper-parameters. The best result was found with the linear SVR model in the case of BXPHARMA with the highest R-squared score of about 97.04%, and lowest RMSE of about 1.23, followed by the KNN regression model with an R-squared score of approximately 96.39% and RMSE of about 1.38. SVR has the lowest RMSE and highest R-squared values in all cases.