Improved PSO_AdaBoost Ensemble Algorithm for Imbalanced Data.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Traditional Adaptive Boosting (AdaBoost) struggles with imbalanced datasets, as it prioritizes misclassified samples over minority class representation.
• Existing AdaBoost methods may generate redundant weak classifiers, consuming excessive system resources.

Purpose of the Study:

• To enhance the AdaBoost algorithm for improved performance on imbalanced datasets.
• To introduce the Area Under Curve (AUC) metric for comprehensive model evaluation.
• To develop an ensemble algorithm that avoids local optima and optimizes weak classifier coefficients.

Main Methods:

• An improved AdaBoost algorithm (AdaBoost-A) was proposed, incorporating the Area Under Curve (AUC) to refine error calculations.
• A novel ensemble algorithm, PSOPD-AdaBoost-A, was developed to optimize AdaBoost by re-initializing parameters and tuning weak classifier coefficients, preventing local optima.

Main Results:

• The proposed AdaBoost-A algorithm demonstrates improved error calculation by integrating misclassification probability and AUC.
• PSOPD-AdaBoost-A effectively processes imbalanced data, showing significant improvements, particularly for datasets with high imbalance ratios.
• The algorithm successfully prevents the generation of redundant weak classifiers, optimizing resource utilization.

Conclusions:

• The PSOPD-AdaBoost-A algorithm offers a robust solution for classification tasks involving imbalanced data.
• Integrating AUC into AdaBoost enhances its ability to handle class imbalance effectively.
• The optimization strategy in PSOPD-AdaBoost-A improves efficiency and classification accuracy for imbalanced learning scenarios.