Artificial intelligence and Artificial Neural Networks (ANN) have gained increasing attention over years in performing various analytical tasks. But the interpretability of results produced by ANN is still limited due to its “black box” nature. This would be problematic in certain fields where explanations of results are required. This inherent disadvantage in ANN has set the stage for fuzzy neural network, a hybrid combination of neural networks and fuzzy systems. This research proposes an evolving fuzzy interpretable recurrent neural network (EF-IRNN) combining accuracy of neural networks and interpretability of fuzzy systems. The proposed EF-IRNN can handle concept drift, where the model can adapt to changes in underlying data distribution and parameters. Fuzzy clustering, feature optimization, and parameter tuning are performed for EF-IRNN. EF-IRNN is applied to financial market data, to predict multi-day look-ahead prices by combining three technical indicators, Moving Average Convergence Divergence, Relative Strength Index, and Rate of Change. The 12 financial assets are employed in the experiments. The EF-IRNN based system is utilized for trading of individual financial assets and multiple assets in portfolios through Reinforcement Learning, to capture trend reversals of prices. It can enhance the traditional backward looking trading indicators to detect trends for stock trading.
