Active sensing can be defined as shaping the sensory feedback signals perceived from the environment via various motor movements. Despite not aiming to serve a motor goal, these movements can enhance the sensory feedback received from the environment by utilizing interaction between the sensory and motor systems. A common natural example to active sensing is the ability of animals to extend their optical sensory volume by moving their eyes, heads and bodies. Our primary goal is to distinguish active sensing movements from the task-dependent movements. Then, we aim to classify active sensing movements based on their characteristics to reveal how these movements assist behavioral task performance. In this context, we will use a dataset that includes the behavioral data collected during the refuge tracking behavior of Eigenmannia virescens, a species of weakly electric fishes. These fish hide between rocks and plants in the water due to their natural refuge seeking behavior. They can track the movements of such refuges via swimming forwards/backwards by benefiting from the long anal fin they developed during the evolutionary process. A similar environment can be achieved in a lab environment by placing a PVC tube into the aquarium. These fish can 'successfully' track the movements of such tubes by swimming forwards/backwards in a single linear dimension. However, in addition to the movements performed for the sake of tracking the refuge, these fish also exhibit some active sensing movements to increase their sensing performance. Detection and classification of such active sensing movements is critical to understand the roles of these movements in sensorimotor control. In this study, we plan to utilize machine learning approaches first to detect the active sensing movements performed by the fish and then to classify them into previously labeled types of active sensing movements.