Target Prophecy in an Underwater Environment Using a KNN Algorithm

Abstract

As detailed in a simulation, the output of localization is directly linked to the data available for machine learning training. Machine learning is a promising tool and can be more efficient and successful than other algorithms if adequate training data are provided. In the absence of an adequate training data set, machine learning might not be a feasible choice. In this chapter, a novel underwater application using the k-nearest neighbors (KNN) algorithm is proposed that aims at pairing machine learning with the acoustic propagation approach, since the underwater application is not tested using KNN because of its large data set. This chapter focuses on the computation of the KNN using a large data set with less computational time. This improves the prediction of the targeted sensor nodes underwater to estimate the sensor node position using the nearest neighbor. From these results, KNN can classify whether the nodes deployed months or years before are active or inactive at present to analyze its accuracy and error of prediction. This emphasizes the fact that inactive nodes can be replaced for a high rate of prediction in lost debris. The error rate is proved to be less in the case of parametric values of k = 8 to k = 15. Experimental results using KNN have proved, with 96.33% accuracy, its use for underwater target prediction for locating the positions of sensor nodes.