Classification of Shark Behaviors using K-Nearest Neighbors

In this paper, we aim to ascertain the behavior of California horn sharks (Heterodontus francisci) based on accelerometer data using a machine learning algorithm called K-Nearest Neighbors. We use digital signal processing techniques such as Fast Fourier Transform (FFT) and Discrete Cosine Transform (DCT) to represent the accelerometer signal in the frequency domain which reduces the data size required for training the classifier, the computation time and the memory resources needed, in addition to improving model accuracy. The shark behavior is classified into four classes namely Resting, Swimming, Feeding and Non-Directed Motion. We compare different combinations of time and frequency domain data on the performance of the algorithm. It is shown that the transform domain data considerably improved the accuracy of the classifier.