Leveraging Genetic Algorithm-Optimized Isolation Forest for Anomaly Detection in Compressor Machines

In modern industries, compressors play a vital role in various applications, including energy production, gas processing, and commercial cooling systems. However, in daily operations, compressors can encounter several issues that threaten operational consistency and efficiency. Common problems include component wear, abnormal vibrations, and leaks. To minimize the risk of serious damage and reduce high maintenance costs, this study proposed an anomaly detection technique using the Isolation Forest algorithm and the application of Shapley Additive exPlanations (SHAP) to identify the most influential attributes in determining compressor anomalies. There are nine main attributes in compressors, including Delivery Pressure, Delivery Temperature, Oil Pressure, Oil Temperature, Horizontal Temperature, Stage 1 Temperature, Stage2 Temperature, Stage3 Temperature, and Water Temperature. To detect anomalies based on these attributes more effectively, the performance of the Isolation Forest was enhanced by applying a Genetic Algorithm to optimize the parameters of the Isolation Forest, which include contamination, n_estimators, max_samples, n_jobs, and max_features. The optimization results from the Genetic Algorithm produced the following Isolation Forest parameters, with contamination at 0.16, n_estimators at 100, max_samples at 267, n_jobs at -1, and max_features at 9. The evaluation results using these optimal parameters showed an average accuracy of 0.993. The evaluation using SHAP calculations demonstrated consistency with conventional methods, ensuring that identifying the most influential attributes in detecting compressor anomalies is accurate and reliable.