Passive sonar signal detection and classification play pivotal roles in underwater surveillance, providing invaluable insights into the aquatic environment and aiding in the identification of submerged objects. This comprehensive paper explores the utilization of machine learning techniques in the context of passive sonar signal classification. We focus our efforts on the ShipsEar dataset, which contains a total of 90 instances representing 11 different types of sonar signals from ships, classified into four distinct classes. To train the models, features such as spectral contrast and chroma were extracted from these instances. Additionally, to enhance the dataset's utility, we have partitioned the original samples into multiple segments, each spanning 10 s. Our work aims to assess the effectiveness of various machine learning algorithms in classifying these ship signals. In this pursuit, we utilize a range of classifiers, including k-nearest neighbor’s (KNN), decision tree (DT), random forest (RT), and logistic regression, with the goal of identifying the most proficient algorithm for ship classification. The results of our analysis reveal that the Random Forest Classifier emerges as the most accurate, achieving an impressive test accuracy of 82%.