Evaluation and Comparison of K-Nearest Neighbors Algorithm Models for Heart Failure Prediction

Abstract

Heart failure is a disease that is one of the most crucial in the world. Researchers have used several machine learning techniques to assist health professionals in the diagnosis of heart failure. K-NN is a technique of supervised learning algorithm that has been successfully used in terms of classification. However, using the K-NN algorithm has stages in terms of data analysis. The data used must also be processed in such a way that it becomes data that is easier to analyse and that the results obtained are also more accurate. Data pre-processing involves transforming raw data into a format that is appropriate for the model. The normalization technique is one of the techniques contained in pre-processing. This research uses two normalization techniques, namely the simple feature scale and min-max. The purpose of this study is to compare the performance of the KNN model to obtain an optimal prediction model. This study contributes to producing a heart failure prediction model based on the K-Nearest Neighbors (KNN) algorithm that can be optimized to improve the accuracy of early detection, so that it can help medical personnel in making more appropriate clinical decisions. The results obtained from this research show that the dataset that uses the min-max normalization method is better than data that is not normalized and data that uses simple feature scale normalization. The highest level of accuracy was achieved by employing the min-max normalisation technique, with a value of K=9, resulting in an accuracy rate of 85.05%.