Boosting Ensemble of CNN for Vision

Abstract

This study investigates methods to enhance the accuracy of Convolutional Neural Net works (CNNs) through the application of Boosting Ensemble methodologies. The re search covers diverse image datasets, including CIFAR-10, CIFAR-100, and Fashion MNIST, aiming to utilize AdaBoost, a widely adopted boosting algorithm, to enhance CNN performance. A crucial aspect of this research is the assessment of AdaBoost’s effectiveness in address ing imbalanced datasets. Imbalanced datasets, marked by uneven class distributions, pose a common challenge in machine learning. Understanding how AdaBoost addresses these scenarios is a central focus of this study. The empirical findings highlight AdaBoost as a valuable complementary strategy for improving CNN accuracy, especially in cases with imbalanced class distributions. An important observation is that the ensemble model, incorporating AdaBoost, achieves a significant 6% higher test accuracy compared to the baseline CNN. This improvement indicates a substantial enhancement in the model’s generalization ability to unseen data.