Machine Learning-Based Framework for Intraday Stock Price Movement Prediction Using Open Prices and Historical Data

The stock market offers potential fortunes but carries inherent risks because one cannot always be on the correct side of the trade. Despite various market research tools, none can guarantee profits, and even professionals recommend using a hard stop loss. This research aims to improve intraday trading strategies through a machine learning framework that uses the day’s open price to predict stock price movements. Key technical novelties include utilizing open price for feature extraction, introducing enhanced technical indicators derived from open prices, integrating multiple time frame exponentially weighted moving average for nuanced trend analysis, comparing parameterized and nonparameterized models to highlight the benefits of hyperparameter tuning, focusing on high-impact Indian stocks, addressing data imbalance and correlation in time-series data, employing a custom chronological data split to preserve temporal dependencies, and using comprehensive performance metrics (accuracy, precision, recall, F1 score) for robust model evaluation. Experiments showed that the extreme gradient boosting model outperformed k-nearest neighbor and random forest models on all datasets. Parameterized models also surpassed baseline models, demonstrating the framework’s effectiveness in delivering timely and accurate trading signals based on open prices and historical data.