Prediction of Underwater Sonar Targets

Abstract: Classification of underwater SONAR returns is necessary for detecting sea mines under the oceans as they impose serious threats to ships and submarines. ‘Connectionist Bench (Sonar, Mines vs. Rocks) Data Set’ chosen from UCI machine learning repository is considered. The dataset is in the form of a CSV file with 60 attributes and 208 records, 111 patterns obtained by bouncing SONAR signals off a metal cylinder at various angles and under various conditions and 97 patterns obtained from rocks under similar conditions. Standardization preprocessing technique is carried out on the data. Standard-scalar utility class is used to generate scaled data. k-Nearest Neighbour and Standard Vector Classifier classification algorithms are used to train the model, evaluate these algorithms and calculate the model accuracy in each case. Principal Component Analysis is performed for feature selection and the models are tuned using optimal hyperparameters to obtain better accuracy. As a result, the Standard Vector Classifier model gives a better accuracy of approx. 93 % compared to the k-Nearest Neighbour model which gives approx. 88%.

Keywords: Machine learning, SONAR, k-Nearest Neighbour, Standard Vector Classifier