Fuzzy-Based, Noise-Resilient, Explainable Algorithm for Regression

The main properties of the presented algorithm are noise resilience and explainability in regression problems. This function approximator has two parts. First, it utilizes a hierarchical clustering approach to initialize its parameters. The information of the clusters is merged with a fuzzy based Takagi-Sugeno-Kang methodology. Second, the model is fine-tuned with gradient descent learning. The last section of the paper is a benchmark with different neural network architectures. The results show an improvement in performance that encourages further use of this technique.