Developing advanced model for portfolio optimization and return prediction using nested adaptive efficient CapsNet with spatial attention

Abstract

The success of selection optimization based on the predicted outcome of stockpile advertises. Recently, innovations in machine learning techniques have highlighted the importance of incorporating predictive theories for portfolio selection. Utilizing return predictions from traditional point series models in portfolio construction improves the effectiveness of actual portfolio optimization strategies. Stocks and bonds optimization is viewed as a multi-objective optimization challenge that has garnered significant interest from researchers, individual investors, and fund managers. A key focus of optimizing the portfolio is to validate the optimal asset weights by improving expected returns along with low risk. An improved portfolio optimization technique establishes an proficient frontier to enables investor on the way to achieve higher estimated returns with minimal threat exposure. Therefore, highly efficient model is required to develop in financial sector. This research combines portfolio optimization and returns prediction to attain the expected returns in the stock market. To be specific, the required stock market-related data are collected from online resources. Then the Nested Adaptive Efficient CapsNet with Spatial Attention (NAECSA) model is incorporated for the return prediction process. The parameters of NAECSA tuned using a Refined Normalized Fitness-based Doctor and Patient Optimization Algorithm (RNF-DPO) to maximize profit and minimize Mean Squared Error (MSE). During the return prediction process, the risk level of the stocks is determined. The same EDPO is used for formulating the optimal portfolio and it is rebalanced by capturing the up-to-date market condition. To estimate the efficiency of the suggested portfolio optimization system in relation to other baseline works using heuristic methodologies, a thorough analysis is conducted.