Analysis of Variants of KNN for Disease Risk Prediction

As we all know, Supervised Machine learning algorithms are gaining a lot of attention these days and are used in various fields. These algorithms play a great role in classifying, handling and predicting the data with the help of machine learning and makes use of a labelled dataset to predict and classify any given unlabeled data. KNN is one of the most popular supervised learning algorithms that delivers excellent results and is widely used these days for predicting the risk of certain diseases. Different variants of KNN are being introduced by different researchers due to distinct inadequacies of the standard KNN. The goal of this research is to evaluate the performance of such KNN variants in disease risk prediction by taking into account various factors such as performance, accuracy and so on. These variants which are already being proposed by different researchers will be reviewed and analyzed in this study. All these variants of K Nearest neighbor are implemented in Python Programming Language on different medical datasets taken from Kaggle. Extensive efforts are being made to analyze the different variants of KNN with respect to disease risk prediction and after implementation, it was noticed that some variants worked really well for some datasets. Out of all, Generalized KNN turned out to be one of the most efficient variants of KNN as it gave a good accuracy of 90.15% for mole cancer dataset. Also, it gave a precision of 90.13%, recall of 94.67% and AUC of 88.61% which overall is a very good performance. It was noticed that the Genetic KNN with GA Algorithm gave an accuracy of 99% for both Chronic Kidney Disease datasets. This study will be really beneficial for other researchers as well in the health industry and will also help in improvising their studies.