HYBRID MACHINE LEARNING MODELS: COMBINING DEEP LEARNING WITH CLASSICAL ALGORITHMS

Abstract

This paper presents a hybrid machine learning framework that integrates deep learning–based feature extraction with classical machine learning classification to enhance image classification performance. The evaluation of the proposed approach is done on the CIFAR-10 benchmark dataset which has 60,000 color images with 10 classes. Two pre-trained convolutional neural network architectures are used to extract deep feature representations and they are ResNet-50 and VGG16. The deep features are then extracted and grouped via a feature fusion strategy and then the combination is then classified with a Support Vector Machine (SVM) and radial basis function (RBF) kernel. The obtained experimental results reveal that the proposed hybrid model attains an accuracy of 94.8% that is superior to that of the standalone CNN model (91.4%) and the classical SVM that is trained on raw pixel features (62.3%). A rigorous cross-validation and statistical significance test of the framework ensures that the framework is robust and has the ability to be generalized. The results suggest that hybrid architectures are effective in integrating deep representation learning and stable decision boundaries that lead to a higher accuracy of classification, a decrease in overfitting, and an increase of robustness. This is a hybrid learning approach that provides a scalable and practical solution to real-world problems involving image classification in which model stability and accuracy are of vital importance.

Keywords: Hybrid Machine Learning, CIFAR-10, Convolutional Neural Networks, Support Vector Machine, Feature Fusion, Image Classification.