Resilient IT Systems: Engineering Approaches for Fault Tolerance and Recovery

Abstract

Reliable banking systems demand uninterrupted service despite failures in infrastructure, network, or software components. Conventional fault tolerance approaches rely on static rules and basic machine learning models, which often fail to capture temporal dependencies, resulting in delayed fault detection and inefficient recovery. The aim is to design a robust, fault-tolerant, and recovery-oriented framework capable of identifying early failure patterns and ensuring rapid system restoration. Novelty is introduced by the Dumbo Octopus algorithm, self-attention with Stacked long short-term memory (DOA-SA-Stacked LSTM), enabling precise temporal learning and adaptive parameter tuning. A generalized dataset comprising system performance metrics, transaction logs, and failure records is utilized. Preprocessing involves the interquartile range (IQR) for data cleaning and the Z-score for normalization to ensure consistency and reliability. Feature extraction is performed using Independent Component Analysis (ICA) to capture temporal variations and system dynamics. The proposed model operates by first learning sequential dependencies through stacked LSTM layers, while the self-attention mechanism emphasizes critical time steps associated with anomalies. The DOA optimizes model parameters to improve convergence and prediction accuracy. Implementation is carried out using a Python-based deep learning (DL) model. Experimental results demonstrate an accuracy of 98.3%, a precision of 98.1%, a recall of 98.6%, an F1 score of 98.4%, and a recovery time of 6.8s, stimulated in Python. The method effectively models failure progression and recovery patterns, ensuring reliable operation in banking environments. The proposed approach provides an efficient and scalable solution for resilient Information Technology (IT) systems, enabling proactive fault management and rapid recovery.