Predictive Maintenance Air Conditioner Using Machine Learning

Abstract

Predictive maintenance will take care of the machine's needs in terms of power loss from damage that lowers performance, operational costs from severe damage, business interruptions from damage that renders the machine unusable, and much more. Almost every home has an air conditioner, the machine that requires constant maintenance of temperature and humidity, especially in offices with servers or control rooms. Preventive and predictive maintenance is necessary to identify the necessary steps for technicians to take when handling an AC before the damage worsens. In this research we implemented and proposed an Air Conditioner detection system using machine learning with three methods, namely K-Nearest Neighbor, Decision Tree, and Random Forest. In order to understand the actual conditions of each AC, we use data sheets that we gathered through surveys with engineering teams at multiple hotels as well as technical teams that handle servers and control rooms. There are 20 features in the gathered data set; however, since only 14 of the features affect the value, extraneous data will be removed. Then the data was divided into two groups, namely 23 AC Failures yes, which means the AC condition is not normal and 110 AC Failures No, which means the AC condition is not damaged. Using the stratified random sample method, 25% of the data will be oversampled. In this study, Kbest and backward elimination were employed for feature selection. The SMOTE approach was then applied for oversampling due to the unbalanced groups. With accuracy values of 91.18%, precision 91.18%, recall 90.90%, and f1-score 90.92%, the Random Forest model with the suggested model outperformed the Decision Tree and KNN models, according to the experimental findings.