Supervised binary image classification using machine learning and convolutional neural networks

No Thumbnail Available
Scanlan, Seamus
Issue Date
Higher Diploma in Science in Data Analytics
Dublin Business School
Items in Esource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
Machine Learning and Deep Learning Algorithms were investigated in terms of their ability to perform a supervised binary image classification task involving the Kaggle Dogs vs Cats dataset. Machine Learning algorithms struggled to achieve above 60% training accuracy. Though the CNNs tended to overfit, the inclusion of regularisation via dropouts reduced this effect and the optimal deep learning algorithm developed using Convolutional Neural Networks achieved a training accuracy of 96% and a validation accuracy using unlabelled images of 94%. In a straight comparison the optimal CNN model had an AUC of 94% compared to 51% for kNN and 58% for Naive Bayes when tested using unseen data.