A Comprehensive Evaluation of Machine Learning-Based Image Classification Models

Abstract

This study automatically categorises photographs using Machine Learning (ML). ML has a feature extractor and classification module. It can retrieve certain picture characteristics but not unique ones for validation. ML automatically identifies authorised application attributes. Many deep learning approaches involve convolutional neural networks. A CNN has input, hidden, and output layers. A matrix of input layer pixel values is reinforced by weights and biases to produce pictures. Convolutional, pooling, or linked obfuscated layers. Fully linked output layers classify images. Another hyper-parameter collection is examined. Parameters substantially impact photo classification. Many studies analyse each hyper parameter’s influence. Adjustments include learning rate, iterations, batch size, filters, and hidden layers. A well supervised machine learning algorithm identifies photos. Feature count affects SVM performance. CNN and SVM are evaluated on massive datasets. CIFAR-10 contains ten classes and sixty thousand pictures; MNIST has seventy. Comparison between Machine learning and SVM. Accuracy, learning, forecasting, and others are metrics. CNN classifies better than SVM. CNN outperforms CIFAR-10 on MNIST. The size, kind, and description of the dataset greatly impact machine learning and deep learning performance. CNN learns and predicts slower than SVM. These findings outperform previous machine learning studies utilising the same datasets and methods.