Dog Breed Classification Using Machine Learning Algorithms

Abstract

The thesis completes a thorough study on the task of dog breed recognition using conventional machine learning techniques on the Orange ML interface. It tackles the problem of recognizing dog breeds by implementing a Logistic Regression, Support Vector Machines, K-Nearest Neighbors, Random Forest, and a shallow Neural Network. The experiments done in this study were successful in classifying breeds which were visually different, but were still struggling with distinguishing breeds which looked alike. This research stands out for proving the effectiveness of no-code machine learning on complex image classification, analyzing the performance of different traditional algorithms, and empirically describing the problems of classification in the image data set. The research serves as a starting point for automated image recognition technology of animals and classifies prospective research areas in fine-grained image classification in the absence of deep learning structures.