Improvement of F-1 Score in Classifying Shark Data into Shark Behaviors

Abstract

The objective of this paper is to improve the F-1 score computed in classifying shark raw-data into behaviors, namely; Resting, Swimming, Feeding, and Non-Directed Motion (NDM). Combining two different sets of pre-processed data into one image is examined for F-1 score improvement. The two sets of pre-processed data are Fast Fourier Transformation (FFT) and Walsh-Hadamard Transformation (WHT). Combining these two sets in a Convolutional Neural Network (CNN) model resulted in considerably improved F-1 score, while combining them in a K-Nearest Neighbors (K-NN) model averaged their individual F-1 scores.