Applying Multiple Linear Regression to Enhance Short-Term Stock Forecasting Accuracy

This research examines the ability of machine learning algorithms in enhancing the accuracy of stock price predictions, a critical aspect of decision-making in the financial sector. Utilizing Multiple Linear Regression (MLR), the research analyzes how the algorithm can be effectively integrated with traditional technical analysis methods such as correlation coefficients and Euclidean distance. This paper presents two simplified computerized approaches using MLR to predict stock prices, with a focus on aiding short-term traders. The models are evaluated using Mean Relative Error (MRE) and Distance Error (DE) metrics. Approaches are evaluated on nine stocks of national stock exchange of India (NSE). Approach I focuses on predicting closing prices using highly positively correlated price segments from historical data, with MLR performing well by leveraging the past seven days of price movements for low-error predictions. Approach II extends this by forecasting High, Low, Opening, and Closing prices, employing the nearest correlated segments from recent data, achieving greater accuracy across all attributes and validating that closely correlated segments enhance prediction. Approach II outperforms approach I across all stocks, with the notable improvements in R ² from 7% to 18%, suggesting both strategies are relatively effective for it. The findings indicate that integrating machine learning techniques with traditional technical analysis tools presents a promising approach for enhancing the reliability of stock market predictions, thereby supporting financial traders in their efforts to achieve profitability.