Enhancing Data Reliability in Cloud Storage Management: Proactive Fault Prevention with Advanced Machine Learning Strategies

Abstract

The increasing reliance on cloud storage systems to manage massive volumes of data has highlighted the need for robust, efficient, and intelligent fault-prevention mechanisms. Traditional fault tolerance methods rely on reactive measures, leading to increased downtime and inefficiencies. This study introduces a proactive fault prevention approach using advanced machine learning techniques to enhance data resilience in cloud storage management. Utilizing the NSL-KDD dataset, undergoes preprocessing, including feature-label separation, one-hot encoding of categorical variables, and feature scaling to standardize input values before being split into training and testing sets. A Long Short-Term Memory (LSTM) model is implemented for failure prediction for binary classification. The model is trained using the Adam optimizer and evaluated with accuracy, loss, confusion matrix, and ROC curve, achieving high accuracy (93.87% training, 92.92% testing). To further enhance reliability, a reinforcement learning (RL)-based replica placement strategy is introduced, dynamically adjusting the number of replicas based on network topology, access patterns, node reliability, and resource constraints. Furthermore, failure to failover provisions allows continuous latency measuring for the system to perform optimally. The integration of resources and the use of predictive analytics in this work greatly improve cloud storage reliability and efficiency.