Optimization-Tuned Hybrid Machine Learning for Imbalanced Big Data Classification

Abstract

Big data classification is a complex challenge, particularly when dealing with imbalanced datasets 
where the minority class is significantly underrepresented. Conventional machine learning algorithms 
often suffer from bias toward the majority class, leading to suboptimal performance. This research 
proposes an advanced optimization-driven hybrid machine learning framework integrating ensemble 
learning and deep learning with a novel multi-objective optimization approach combining Particle 
Swarm Optimization (PSO), Genetic Algorithm (GA), and Multi-Objective Evolutionary Algorithms 
(MOEAs) for hyperparameter tuning. The framework employs advanced feature engineering 
techniques, enhanced data resampling methods such as ADASYN and NearMiss, and explores deep 
learning integration with hybrid CNN-LSTM models. The performance is evaluated on multiple 
benchmark datasets using metrics such as Precision, Recall, F1-score, and AUC-ROC. Experimental 
results indicate that the proposed methodology significantly improves minority class detection and 
achieves higher accuracy compared to existing techniques. The findings provide a robust and scalable 
solution for imbalanced big data classification.