Diabetes Data Classification using Deep Learning Approach and Feature Selection based on Genetic

Abstract

The area of deep learning research (DL) has seen significant growth in recent years and has shown incredible promise for AI in medical applications. In this approach, we have proposed a deep neural network framework for the categorization of feature-based diabetes data. The dataset is classified using the SoftMax layer after features are collected from it using the wrapper approach. Additionally, the network is adjusted through supervised grid research with the training dataset. However, due to the risk of misdiagnosis in a medical context, we evaluated our model using measures such as precision, recall, specificity, and F1 score and obtained superior results. The proposed architecture has been tested using Diabetes 130-US hospital data, which consists of 100,000 patient files with an average of 55 attributes. As well, five machine learning algorithms—Nave Bayes, Random Forests, Decision Trees, Support vector machines, and Ensemble learning—are used to compare results from experiments. The classification accuracy of the ML algorithm was 86%. This research suggests a Deep Learning system based on feature selection for classifying diabetes data. This method has been tested on UCI machine learning data and has outperformed several other classification techniques, reaching 85.61% accuracy in this study.