A comparative study of Activation functions for Diabetes detection using Convolution Neural Networks (CNN)

Abstract

Diabetes is a matter of concern for the health of the entire world, its diagnosis and cure are among the prime challenge for the medical fraternity, because it can be controlled but can’t be cured, sooner the diagnosis the better it will be for the patient. The latest advancements in the field of Artificial intelligence (AI) found that Machine learning (ML) and Deep Learning (DL) approaches are quite effective in predicting the diabetes, and helps in controlling Blood Glucose (BG) levels.Deep learning is a budding field in predictive analysis and is often used in health care applications where the prediction of diabetes is required to be identified in an early stage. The technique Convolution Neural Networks (CNN) of one of the most promising concept, use for prediction purpose.The performed study compares the performance of CNN model on the basis of the metrics obtained after using four activation functions viz. ReLU, eLU, tanh, and Sigmoid, over UVA/Padova dataset from UCI machine learning repository. The performance of CNN using different activation functions is measured based on various statistical measures such as recall, precision, F-score and accuracy. The experimental results show that CNN gives the maximum accuracy of 97.3% when applied with the eLU activation function.