A Heterogeneous Ensemble Learning Approach for Leather Species Prediction

This paper proposes heterogeneous ensemble learning for automatic leather species identification. I t combines the advantage of four learning algorithms for reliable and robust results. The objective is to predict the leather species from novel digital microscopic leather images. Hence, firstly, it extracts seventeen derived and 512 deep features to describe the species-definite leather image behavior. Secondly, these features are used to train the four base learners: k-nearest neighbor (KNN), support vector machine (SVM), decision tree (DT), and Multi-Layer Perceptron (MLP). Thirdly, an ensemble model is built by fusing the prediction accuracies of four models in a voting scheme. Lastly, it utilizes data partitioning to achieve better generalization. The experimental analysis affirms that the proposed ensemble model outperforms the base learners with 98.75% accuracy and 98.77% precision. It resulted in robust results with negligible misclassification (1.25 %). Thus, the present work sufficiently l earns divergence from images, features, and models. Therefore, ensemble learning is appropriate for learning complex leather image data.