Timely Diabetes Possibility Prediction Using AI Techniques

Abstract

The fastest chronic life-threatening disease affecting more than 422 million people globally is diabetes. The primary causes of type 2 diabetes are lifestyle choices and environmental factors. It is a slowgrowing disease which starts to develop metabolic factors long before they evolve into a disease and is formally diagnosed by a fasting sugar test. There are records of indications related to diabetes from 520 patients in the dataset that was used. It contains data on people, such as age, sex and symptoms that may lead to diabetes. I analysed the dataset using Naive Bayes Classifier (NB), Logistic Regression Classifier (LR), J48 Algorithm, Random Forest (RF) and Multi-Layer Perceptron (MLP) Algorithm. After applying ten-fold Cross-Validation and Percentage Split evaluation techniques to this dataset, MLP was found to have the best accuracy. MLP achieved 98% accuracy with this dataset and very few numbers of misclassification. In the medical field, dealing with ambiguous and uncertain data is also a significant concern. In recent years, managing ambiguity in medical data has received more attention. The adaptive neuro-fuzzy inference system (ANFIS) was used in this work to identify diabetes. It combined fuzzy logic's learning capabilities with neural networks to describe uncertainty in expressiveness. To represent uncertain circumstances, fuzzy logic is used, and the model is trained by a neural network. The neural network of ANFIS is based on mathematical computations and is linked with the Takagi–Sugeno fuzzy inference system to tackle complicated problems. The Pima Indian Diabetes Dataset (PIDD) was trained and tested for classification using MATLAB. I have utilised this method to diagnose diabetes by leveraging its great uncertainty-handling capabilities and interpretability to produce good classification results.