Adaptive Self-Evolving Machine Learning Models Using Meta-Optimization for Dynamic Data Environments

Abstract

Smart cities, healthcare monitoring systems, financial markets, and IoT platforms are examples of dynamic data environments where data distributions vary over time. When ideas change, traditional machine learning models that were trained with static assumptions don't always do well. To evolve models autonomously in dynamic environments, we provide a new framework called Adaptive Self-Evolving Machine Learning (ASE-ML). This framework employs meta-optimization and constant learning. Model architecture, hyper parameters, and learning approaches are dynamically modified by a meta-learner powered by real-time performance feedback and drift indicators. Extensive testing on benchmark dynamic datasets shows that the proposed method outperforms standard retraining and incremental learning approaches in terms of computational efficiency, stability, accuracy, and adaptability. When applied to real-world, dynamic data streams, the results demonstrate that ASE-ML can be a dependable choice for next-generation intelligent systems The paper currently presents results mainly through comparative figures and discussions but does not provide exact numerical values (e.g., accuracy = 94.2%, F1-score = 92.8%)