AN ADAPTIVE FRAMEWORK FOR MITIGATING JOB FAILURES IN CLOUD COMPUTING VIA MACHINE LEARNING AND DYNAMIC RESOURCE MANAGEMENT

Cloud systems are increasingly challenged by job failures caused by fluctuating workloads and unpredictable resource availability. While much of the existing research emphasizes failure prediction, less attention has been given to implementing mitigation strategies post-prediction. This paper introduces an Adaptive Failure Mitigation Framework that combines machine learning-based failure prediction with adaptive resource management techniques. Using historical logs and system metrics, the framework applies the Random Forest algorithm to classify potential failure types. Based on these predictions, a rule-based Cloud Manager dynamically executes appropriate corrective actions tailored to job priority, predicted failure category, and real-time node availability. The corrective strategies include dynamic resource reallocation, job rescheduling, live migration to alternative servers, and replication of critical tasks to ensure continuity and data safety. Evaluations conducted using the Google 2019 dataset show the framework achieves a high classification accuracy of 99.98% in identifying failure types and successfully mitigates 91.2% of predicted job failures.