An Explainable ML Model for MultiClass Marine Vessel Classification Using Passive Sonar Signals

Detection of vessels navigating through water is a critical aspect of maritime surveillance, with passive sonar data playing a pivotal role. It not only aids in underwater monitoring but also contributes significantly to identifying and tracking vessels moving through water bodies. This work delves into leveraging machine learning methodologies for effectively categorizing passive sonar signals, centering on the utilization of the ShipsEar dataset. To bolster the dataset's utility, the original samples are segmented into 10-second intervals. From these segments the following features like spectral contrast, chroma, zero-crossing rate, tonnetz, and mel-frequency cepstral coefficients (MFCCs) were extracted. Our primary goal is to evaluate the efficacy of diverse machine learning algorithms in accurately classifying sonar signals. To achieve this, we employ classifiers such as k-nearest neighbors (KNN), decision tree (DT), random forest (RF), and logistic regression (LR). Finally, random forest classifier shows an efficient performance achieving an impressive accuracy of 0.94. Furthermore, the study integrates explainable artificial intelligence (XAI) techniques, notably local interpretable model-agnostic explanations (LIME), to delve deeper into the classification process, unraveling profound insights into the complex domain of passive sonar signal classification.