Animals’ Classification: A Review of Different Machine Learning Classifiers

Abstract

Abstract

Animal classification has recently attracted wide interest from ecologist. There have been attempts in the literature to apply image recognition methods to classify animals. The diversity in animal species with their intricate intra-class variability and interclass similarities cannot be accurately represented by these existing algorithms, despite their promising results for image recognition. This article strives to classify animals based on their different unique attributes, rather than using image recognition. Accordingly, the article evaluates the classification abilities of a few machine learning (ML) tools, including support vector machines (SVM), Knearest neighbours (KNN), and decision trees (random forest (RF) and J48). The result was verified using the dataset taken from Irvine machine learning repository (University of California), which consists of 108 animals with 18 attributes. Besides, the performance of these ML tools was documented for different experimental conditions in terms of their classification accuracy (sensitivity) and classifier reliability (false discovery rate). The SVM classifier exhibits better false discovery rate and classification accuracy performance as compared to the KNN, J48, and RF classifiers. Yet, all of these ML tools can be deployed for real-time animal classification depending on end-user application requirements and formulations.