Cuckoo Search Algorithm Features based Multi Kernel Support Vector Machine Performance Analysis for Detection of Battery Anomaly from Lithium-Ion Battery

An important factor in assuring system reliability and safety is the detection of anomalies in lithium-ion battery voltage in a timely manner using machine learning techniques. In this study, the NASA battery voltage dataset is first preprocessed with a One-Class SVM to remove the outliers. Following this, the Cuckoo Search algorithm was used in feature extraction to select the most informative parameters for classification. Then, the extracted features are evaluated using Support Vector Machine classifiers through Linear, RBF, and Polynomial kernels. Among them, the RBF kernel achieved the highest classification accuracy of 96.87% and MSE at 0.0741, thus effectively performing anomaly detection in the battery voltage profiles.