Stock Portfolio Selection with Supervised Classification

This study employs a supervised learning approach for the Stock Portfolio Selection (SPS) problem. The proposed approach solves three problems defined in the literature review: the narrow period selection problem, the proper objective selection problem, and the comparable peer selection problem. Three classification methods are utilized to categorize the stocks into select (purchase) and ignore (do nothing) classes. The genetic algorithm is used for training the classification parameters. Trained individuals together form a portfolio of stocks based on a voting mechanism. Forty-quarters of the data of 29 of 30 DJI index stocks are chosen for experiments. Statistical analysis of experimental results shows that focusing on the defined problems helps in choosing a portfolio of stocks that beats the market.