Unlocking the Potential of Machine Learning in Portfolio Selection: A Hybrid Approach with Genetic Optimization

In the field of financial market predictions, machine learning has been widely used to identify patterns and gain valuable insights. However, for success in portfolio selection, it is crucial to optimize factors that impact accuracy. This study focuses on combining machine learning and optimization to enhance stock selection and prediction capabilities. The work starts with hyperparameter optimization and utilizes three different machine learning algorithms: XGBoost, LSTM, and Deep RankNet. Our findings show a 40% improvement in results through the use of a genetic-based optimization technique, as well as a promising daily average return of 0.47% through a novel feature engineering approach. The study provides a framework for optimizing and learning in financial portfolio selection, with promising results for medium and small-sized traders.