The growth of data-driven systems in non-stationary environments necessitates intelligent models capable of continuous adaptation and efficient optimization. Conventional static machine learning pipelines fail under distributional shifts, while purely adaptive models often suffer from instability and high computational cost. This paper proposes a Dynamic and Intelligent Data Science Framework that integrates continual learning, automated hyperparameter optimization, and resource-aware model reconfiguration within a unified architecture. The proposed framework enables real-time or near-real-time adaptation through incremental updates combined with optimization-driven control mechanisms. Experimental results demonstrate significant improvements over baseline approaches, including an adaptability gain of +5.8% compared to +1.9% (online learning) and -5.1% (static models), improved robustness (0.92 vs 0.85 and 0.77), and competitive computational efficiency (1.15× vs 0.95× AutoML and 0.80× static models). Additionally, the framework achieves a balanced stability–plasticity trade-off, ensuring sustained performance in dynamic environments. These results validate the effectiveness of integrating adaptive learning with optimization-aware strategies for building robust, scalable, and intelligent data science systems.
Dalela et al. (Fri,) studied this question.