Explainable AI-empowered Neuromorphic Computing Framework for Consumer Healthcare

Abstract

Machine Learning has evolved significantly over the last decade, with models capable of robust and accurate predictions over data of various categories. The impact is rather big in the field of Healthcare. However, deployment of small, energy-efficient models remains a goal pursued by researchers across the field. Futhermore, the model inferences are hard to interpret and draw conclusions from, especially in healthcare, where its very important to know how the diagnosis decision was passed and what factors are causing the issues. In this paper, we address both of these shortcomings, where we develop a neuromorphic computing based machine learning model, namely an SNN, and using SHAP, LIME, and eli5 explainability techniques to explain the predictions of the Spiking Neural Network. The proposed SNN performs at an accuracy of 85.06%, better than the deep neural network by 5.73% for the Diabetes dataset and at an accuracy of 97.83% for the Mobile Health dataset, thereby mitigating the issue of both lack of performance-cum-efficiency, as well as uninterpretability of these machine learning models.