Comparative Analysis of Machine Learning Algorithms for Animal Species Classification Using the Zoo Dataset

Abstract

Abstract: Artificial Intelligence (AI) is an important instrument for biologists, as it can be directly applied to the classification and analysis of complex biological data. The precise species identification of animals is vital for conservation of biodiversity, ecological surveillance, wildlife management, and biological studies. Conventional animal classification methods need rigorous professional knowledge and an animal's biological features were observed by the human mess which is a time-consuming process and would be bias or error with different observers. In this work, we compare the performance of four supervised learning algorithms, Decision Tree, Random Forest, K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) for plant species classification based on the well-known publicly available Zoo dataset. The dataset has a total of 101 animals working by sixteen attributes and they are divided into seven different classes. Data pre-processing involved removal of non-informative data and formatting of feature set for the model-training. The results of each classifier were tested using 10-fold cross validation on 80:20 train-test split results and validated with 10-fold cross validation. Standard evaluation procedure such as accuracy, precision, recall, F1-measure was used to evaluate model performance and confusion matrix and feature importance was used to interpret the behavior of classification. The experimental results demonstrate that the Decision Tree and the Random Forest classifiers obtained 100% accuracy on the test dataset while KNN and SVM obtained accuracy. Cross validation also yielded similar results with Decision Tree and Random Forest having a mean accuracy of 96.00% which confirms robustness of proposed method. The feature importance analysis of biological features including feathers, milk, eggs, legs, etc. had the highest contribution to species classification accuracy. Results show that bagging and tree-based learning techniques perform very well for the structured biological data, and they can be used to develop intelligent animal classifiers, as well as to analyze and interpret ecological data.