A systematic evaluation of LSTM, XGBOOST and hybrid models for intraday trading using a risk-focused machine learning framework

Abstract The stock market is an important component of personal finance and presents many investment instruments, including equities, mutual funds, and trading. Growing awareness and the availability of technology have led more people to start trading with the view to realizing significant financial gains and security over the long run. In the era of advanced technologies, the financial market is no exception to their influence. The main goal of the study is to disprove the popular myth that high predictive accuracy is the only reason behind high returns. Rather, it highlights how the structure and execution of an effective trading infrastructure is more important in delivery of a stable financial performance. The purpose of this paper is to develop a systematic improvement and comparative analysis which would help to evolve a simple predictive model with a standard accuracy of 0.53 into a strong framework with the ability to attain 10X financial goals. The performance gains seen can be attributed to the systematic combination of quantitative risk management techniques, which include algorithmically specified stop-loss limits and position sizing rules, and the compounding effect that causes the capital to grow exponentially within the trading structure. The scientific contribution of this study lies in bridging predictive modeling and trading system design, demonstrating that the integration of risk management, transaction costs, and compounding dynamics plays a decisive role in realized trading performance, beyond predictive accuracy alone. The experimental framework setup consists of four selected models: LSTM, XGBOOST, an integrated XGBOOST–LSTM model, and XGBOOST augmented with moving average features.