Classification of Diabetes using the Biologically Plausible Network of Spiking Neurons

Abstract

Diabetes is a disease that is becoming more common everywhere every day. The increased blood sugar level caused it to start. This sort of illness may be chronic and manifest when the patient's body is unable to produce enough insulin or fails to use it. The numerous computer or electronic-based systems were identified using various categorisation strategies for predicting and diagnosing diabetes. The “third generation” of the neural network, i.e., the Spiking Neural Network (SNN), is getting much attention in the research community due to the akin dynamics of biological neurons, energy efficiency, and computational power. SNN is the future for bridging the gap between neuroscience and the Machine Learning (ML) community. The power of SNN can be beneficial in classifying diseases in medical science. This paper's experimentation is done on automatic diabetes prediction. We have observed that the SNN produces significantly superior results in terms of accuracy, computational cost and biological plausibility than that of the state-of-the-arts through careful experimentation on the Pima Indian Diabetes dataset drawn from the Kaggle. The proposed model of SNN uses the SpiFoG learning algorithm, an efficient, biologically plausible supervised learning algorithm, to train SNN models using a metaheuristic approach that uses genetic algorithm (elitist floating-point type) with a hybrid crossover along with the dynamics of the neuron model, and random delay in the synapse model.