Selection and multi-objective optimisation of stock portfolio using a combination of machine learning methods and meta-heuristic algorithms

The main goal the model and optimal investment portfolio selection to maximize stock portfolio returns based on the forecasted price and minimize investment portfolio risk based on the Markowitz model. This paper presents is to select the optimal stock portfolio based on data training through Markov decision-making and ensemble learning. To teach data from the data of five years (2016-2011), 85 active stock exchange companies in Iran that have been filtered based on technical, fundamental, and time series variables have been used. Therefore, the stock sets are first filtered based on optimizing trading rules based on technical analysis, Markov decision-making and ensemble learning that issued the buy signal. Data for the next 5 years (2020-2016) were also used to test NSGA II and MOPSO algorithms. According to the obtained results, if the shares are bought equally among 85 companies and maintained for five years, the average return on the total stock portfolio is equal to 13.08%, with a risk of 0.946%. While using the MOPSO algorithm has achieved an average of 43.54% with an average risk of 1.102% . The rate of return on capital for the NSGA II algorithm was also the highest in 5 years. Therefore, it can be said that based on the obtained indicators, NSGA II algorithm is the best combination of the stock portfolio.