Fuzzy adaptive learning control network with policy and another adaptive resonance theory (FALCON-PAART) embedded deep structure with applications in stock market prediction and analysis

Abstract

Deep learning has made breakthrough in the field of computer science. It is known for its strong ability to learn the input data’s increasingly abstract representations and has shown promising results in many areas, such as Healthcare to detect abnormality, Finance to automate trading. This paper explores the learning capabilities of deep learning by embedding a Deep Learning Model within a Fuzzy Adaptive Learning Control Network with Policy and Another Adaptive Resonance Theory (FALCON-PAART) Model. FALCON-PAART Model focuses on predicting the turning points in the stock market through the usage of Adaptive Resonance Theory (ART) to build fuzzy input and output clusters, complementary learning to segregate positive and negative knowledge and momentum policy to serve as a memory for the previous states of the market. The embedding of the deep learning model the FALCON-PAART architecture allows the data-driven fuzzy implication to provide a close correspondence to real-world entailment of data, rendering the entire framework explainable for each individual prediction. This is because both the deep structure and the fuzzy structure share a common input and output linguistic, hence, we are able to associate the inference process of the RNN with fuzzy rules. The effectiveness of the proposed model is evaluated on three financial stock prediction datasets of low correlation. Experimental results showed that the proposed model is able to consistently outperform other momentum indicators and trading strategy in all three stocks. Overall, the Deep FALCON-PAART Model has combined strong interpretability and learning capability to serve as an insightful model to make informed decision for Stock Trading.