A HYBRID MACHINE LEARNING METHOD FOR IMAGE CLASSIFICATION
Deep learning (DL) technologies are currently a trendy topic because they aim to understand concepts more precisely by analyzing data at a high level of abstraction through non-linear understanding. This enables them to achieve high performance in image classification, especially for tasks related to medical diagnoses, like analyzing images of the brain’s histopathology. Neural networks (CNN) are the most used deep learning models for diagnosing and analyzing medical imaging data. However, CNNs have some drawbacks, such as requiring a large amount of training data, being prone to overfitting, and having many hyperparameters that need to be tuned. Hyperparameter optimization is the process of finding the optimal values for these parameters that control the behavior and performance of the model. CNN must be implemented at a significant computational expense, and various parameters may need to be adjusted. In this study, we propose a hybrid strategy by fitting the VGG16 pre-trained model with the LSVM classifier for classification of a brain image as normal or abnormal. The VGG16 model is a popular CNN architecture that has been trained on a large dataset of natural images and can extract high-level features from any input image. The LSVM classifier is a linear support vector machine that can learn a decision boundary between two classes of data. By combining these two methods, we aim to reduce the computational cost and improve the accuracy of the classification task. An accuracy of 98.24% is the demonstrated performance using this proposed method on the test set. It was found to be better in terms of accuracy, error rate, sensitivity, F-1 score, and specificity, according to the experimental results. This shows that our hybrid strategy is effective and robust for brain image classification.
Deep Learning, Convolutional Neural Networks, VGG16, Image Classification, Linear Support Vector Machine.