Risk based portfolio reallocation using interpretable fuzzy deep neural network (IFDNN)

Abstract

With the advancement of deep learning, the artificial neural network has become a state-of-theart technique used in almost every problem to provide a robust and efficient solution. These neural networks outperform machine learning techniques in solving complex problems by understanding the data and extracting meaningful features that can help make accurate predictions for the future. The deep learning model learns the intricate patterns from past data during the training process to make better decisions for businesses. However, the problem with these networks is the lack of interpretability in these models, because the end user cannot understand the reasoning behind the decision. Despite the models giving highly accurate results in every field, it does not work well in tasks where external interference is involved. On the other hand, a fuzzy logic system based on if-then rules provides interpretability and lacks in learning from data. That’s where the need for fuzzy neural networks arises to solve the problems that involve human interference. Financial management is crucial in the modern world for investors to decide which stocks to invest assets. This project proposes an Interpretable Fuzzy Deep Neural Network model to predict risk-based portfolio allocation, that assists investors in deciding where to invest their assets in diversified stocks for high returns. Analyze the stock markets in different periods to understand the concept drift, then the IFDNN model is utilized to forecast the multiple lookahead to detect trends. Evaluating the dynamic portfolio allocation and rebalancing improves the momentum indicators.