Interval Type-2 Fuzzy Logic Based Stacked Autoencoder Deep Neural Network For Generating Explainable AI Models in Workforce Optimization

Abstract

In Utility based industries that employ a large mobile workforce, efficient utilization of field engineers is key to optimal service delivery. The utilization of the engineers can be improved by predicting the future performance of work areas by using machine learning tools such as Deep Neural Networks (DNNs).The dramatic success of DNNs has led to an explosion of its applications. However, the effectiveness of DNNs can be limited by the inability to explain how the models arrived at their predictions.In this paper, we present a novel Type-2 Fuzzy Logic System (FLS) whose inputs are preprocessed by a Stacked Autoencoder Neural Network to add some interpretability to a Deep Neural Network model. The proposed type-2 FLS will contain a small rule set with a small number of antecedents per rule to maximize the model's interpretability. We also present an algorithm which can be used to efficiently train the proposed model.We will compare the proposed model with a Standard Stacked Autoencoder Deep Neural Network, a Multi-Layer Perceptron (MLP) neural network and an Interval Type-2 Fuzzy Logic System.The results show that even though the Standard Stacked Autoencoder and MLP Neural Networks have better performance, they do not provide any insight into the reasoning behind the predictions. The Proposed model, on the other hand, provides better result than the standalone type-2 FLS and a comparable performance to the neural networks and provides a little bit of insight into the decision-making process. Without this insight, we cannot be sure why there is a drop in the performance and we need to further analyze the WA before we can take any decision. This leads to quicker decision making and potentially improving the efficiency of the engineers.