Improving the Efficacy of Adaptive Boosting Classifiers by Dynamic Weight Adjustment

     The AdaBoost algorithm has proven robust performance across many classification possibilities. However, its applicability in real-time is frequently restricted by its susceptibility to noisy, imbalanced, and high-dimensional data. This study proposed a Dynamic Weight Adjustment AdaBoost (DWA-AdaBoost) to improve AdaBoost’s classification efficiency in such scenarios. The approach dynamically adjusted sample weights based on classification difficulty and an estimation of local noise. It integrated a noise-sensitivity element into the weight adjustment process, enabling the system to dynamically decrease the impact of noisy or misclassified instances. The methodology included performance measurements and computational efficiency using MATLAB simulation after cleaning, processing all datasets, and utilizing a 5-fold cross-validation approach to provide a more dependable evaluation of the performance of the models. The method’s effectiveness was assessed using three real-world datasets: Spambase, Wine Quality, and Diabetes. Empirical evaluations showed that DWA-AdaBoost continually surpassed standard AdaBoost across many essential performance metrics. On the Spambase dataset, the model achieved gains of 3.04% accuracy, 4.85% precision, and 3.89% F1-score. On the Wine Quality data, the model achieved gains of 2.4% accuracy, 1.19% precision, and 0.63% F1-score. The results demonstrated the algorithm’s resilience to noise and fast convergence, emphasizing its suitability for real-time predictive analytics in spam detection, healthcare diagnostics, and quality control applications.