Analysing Performance and Accuracy of Diabetes Prediction using Hybrid Machine Learning Models

Diabetes is a common and widespread ailment brought on by either an inadequate amount of insulin or improper use of the pancreatic hormone. Although it typically affects middle-aged and older adults, it can also be diagnosed in younger individuals for hereditary or life-related reasons. Health issues result from an increase in blood glucose levels. Although diabetes cannot be cured, early and accurate diagnosis enables lifestyle modifications and preventative measures to greatly lessen the severe health issues linked to the disease. However, the inability of labelled datasets to adequately capture all outliers makes it difficult to diagnose diabetes directly. The National Institute of Diabetes and Digestive and Order Conditions' PIMA Indian diabetes dataset is used in this work. Age, blood pressure, insulin, glucose, BMI, and other colorful features are used to predict the presence of diabetes in certain situations. The unbalanced data is also pre-processed and EDA'd. To deal with the unbalanced dataset, a type of median inference is developed. Several machine literacy techniques are employed for the dataset's validation and analysis, and their performance standards are met in order to produce the most reliable predictions. Both separately and in combination, machine learning techniques such as K-Nearest Neighbours (KNN) and LightGBM are used to generate the predictions.