Web application based Diabetes prediction using Machine Learning

Abstract

Diabetes is a worldwide epidemic that affects millions of people. Long-term consequences, such as cardiovascular disease and renal failure, are more likely to occur in people with diabetes. If this condition could be diagnosed at an early stage, people may live longer and better lives. Diabetic primary care can benefit from many supervised machine learning models educated on relevant datasets. Finding reliable classifier models for diabetes detection using clinical data is the focus of this investigation. This article introduces a number of machine learning techniques, such as the decision-tree (DT), naïve-bayes (NB), k-nearest neighbour (KNN), random-forest (RF), gradient-boosting (GB), logistic-regression (LR), and support vector machine (SVM) that may be taught using a variety of datasets. We have used effective preprocessing methods, such as label-encoding and normalisation, to raise the quality of the models' predictions. Additionally, we have isolated and prioritised a variety of risk variables utilising different feature selection methods. Extensive tests have been run on two separate datasets to evaluate the model's efficacy. When compared to other previous studies, our model shows improved accuracy, ranging from 3.71 percent to 15.13 percent, based on the dataset and the ML technique used. At last, a machine learning algorithm with the best accuracy is chosen for research and development. We use the python flask web development framework to incorporate this model into a web application. This study's findings provide preliminary evidence that using a suitable preprocessing pipeline on clinical data and using ML-based classification might improve the accuracy and efficiency of diabetes prediction.