Frequency Domain Transformations and CNNs to Predict Unlabeled Shark Behavior With GPS Data

Introduction

Recently, data science and machine learning (ML) are applied to many different areas as more disciplines contribute to consider, understand, explain, integrate, and define what learning is from models as well as algorithms. For example, ML can be used to analyze raw data and assist us to make conclusions and can be applied for animal behavior classification as described in the work of Williams et al. (2017), Whitmore et al. (2016), Adams et al. (2019), and Glass (2017). ML can be integrated within an automated process with algorithms that work with raw data for fault circuits, classification in microgrids, demonstrated by Karan (2020), or engineering applications as demonstrated by Um (2017) and Ali et al. (2021), manufacturing for goods production as shown by Wuest (2016). For this multidisciplinary topic (biology, digital signal processing, ML, telecommunication, and GPS), the authors organize this paper as follows. In the next section, the background knowledge on neural networks and theory are discussed. Then, in the main section, the shark behavior, shark data, pre-processing techniques, and GPS data are presented. Finally, the results and overall performance of the CNN architecture based on the frequency domain algorithm is compared with that of the time domain approach.