The increasing complexity of real-world data requires machine learning systems that are not only precise but also adaptable, scalable, and transparent. This study suggests a strong, modular ML pipeline that can help AI architectures that can update themselves by using adaptive hybrid classification models. The framework starts with getting a dataset of network traffic and doing some exploratory analysis on it. Then, it moves on to standardisation and domain-specific feature engineering. To make sure the evaluation is fair, the data is split into training and testing sets, and a variety of base classifiers are trained. These models are then put together in a stacking ensemble with a meta-learner that smartly combines their outputs, using their strengths to make up for their weaknesses. Using standard metrics like accuracy, precision, recall, F1-score, and ROC-AUC, we look at the model's performance. The study employs XA techniques, specifically SHAP to improve transparency and trust by offering an interpretable perspective on feature contributions in real-time packet classification. The findings validate that the optimised stacking ensemble markedly surpasses individual models, offering a dependable and comprehensible method for fault detection. This work not only shows how useful hybrid classification and continuous learning are, but it also sets the stage for AI solutions that can grow and change on their own and can be used in important areas like cybersecurity and real-time anomaly detection.
Sharma et al. (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: