OPTIMIZING IMAGE CLASSIFICATION MODELS: A STUDY ON COMPUTATIONAL EFFICIENCY AND ACCURACY

Abstract

In computer vision, image classification remains a challenging and critical routine, widely applied in domains such as healthcare, security, and autonomous systems. This study evaluates the efficiency of various algorithms implemented for image classification using machine learning, including traditional models such as Decision Tree (DT), Random Forest (RF), k-Nearest Neighbors (k-NN), Support Vector Machines (SVM) alongside deep learning models such as Convolutional Neural Networks (CNNs) and ResNet. The research adopts an experimental approach using a dataset containing images of animals (cats, dogs, and pandas), processed and classified using different algorithms. Performance was assessed using F1-score, recall, accuracy and precision. The study reveal that deep learning models significantly outperform traditional machine learning techniques, with a pre-trained CNN achieving the highest accuracy (91.4%), followed closely by ResNet (89.6%) and Random Forest (88.2%). Among traditional classifiers, SVM and Decision Trees performed competitively with 85.6% and 85.4% accuracy, respectively, while k-NN and Naïve Bayes achieved lower results. Furthermore, ensemble methods demonstrated superior performance, with XGBoost achieving the highest accuracy (96.8%), outperforming Random Forest (95.5%) and boosting techniques like AdaBoost (92.4%). The study concludes that CNNs and ensemble methods provide the best results for complex image classification tasks, whereas traditional algorithms remain viable for smaller datasets with computational constraints. These findings serve as a guide for selecting optimal machine learning techniques for various image classification applications.