Autonomous Self-Evolving Automl Framework using Reinforcement-Guided Neural Architecture Search and Concept Drift Adaptation

Artificial intelligence systems operating in real- world environments must adapt to continuously evolving data dis- tributions. Traditional machine learning pipelines require manual architecture design and hyperparameter tuning, which limits scalability and adaptability. This paper proposes an autonomous self-evolving AutoML framework that integrates neural architec- ture search, reinforcement-guided evolution, transformer-based predictive modeling, concept drift detection, and self-healing mechanisms.The proposed system maintains a population of neural archi- tectures and continuously evaluates them on streaming data. A reinforcement controller guides architecture mutation and model selection, enabling the system to automatically discover improved models over time. Concept drift detection mechanisms monitor changes in the data distribution and trigger adaptive retraining when significant drift occurs. Experimental evaluation using simulated system monitoring data demonstrates that the framework successfully maintains predictive accuracy while autonomously evolving model architec- tures. The results show improved performance across multiple evaluation metrics including mean squared error, R squared score, precision, recall, F1 score, and accuracy. The proposed framework represents a step toward au- tonomous artificial intelligence systems capable of continuous self-optimization and adaptive learning.