Optimization of K Value at K Nearest Neighbor for Classification and Prediction of Healing in Covid-19 Patient

Abstract

In March 2020 WHO declared the coronavirus outbreak that causes Covid-19 a global pandemic. This disease is an infectious disease caused by the SARS-CoV-2 virus which causes mild to moderate respiratory infections and recovers without requiring special treatment. However, some will become seriously ill and require medical attention. The use of information technology in data science and machine learning can help in the fight against this pandemic, one of which is by creating a method that can classify and predict the recovery period of Covid-19 patients. However, there is no symptom-based model to predict the recovery period of Covid-19 patients that can improve clinical decision-making and become an alternative for resource allocation for treating patients in hospitals. Here we propose to test a symptom-based model to classify and predict the recovery period of Covid-19 patients using the KNN algorithm. This algorithm is a simple algorithm and has been widely used in various fields. This algorithm works by classifying an object into a class based on the neighboring distance of the object. The experiment began with data cleaning of Covid-19 patient data, then the classification process was carried out using the KNN algorithm. The test is carried out with and without optimization on the value of k. The first test without optimization with a default value of k=5 obtained an accuracy value of0.77%, while testing by optimizing the value of k with Grid Search CV obtained an accuracy level of0.86% with a value of k=1. From the test results, it can be seen that optimization on the value of k can increase the level of accuracy by 0.09%. For the prediction of the test results are displayed in the confusion matrix. This research will only focus on efforts to predict the recovery period of Covid-19 patients based on medical record data for Covid-19 patients in Pekanbaru, Indonesia.