An Information Technology (IT) equipment fault prediction system for an it support unit

Abstract

A common expectation for high-end customers in most system operated environment is that systems must never fail, hence activities should be seamless. Information Technology components (hardware and software) are inherently prone to failure.

Though these systems rarely “crack”, hardware components can still fail causing software running on them to fail as well. These fault situations have high cost to management and to customers who may experience poor service. If a fault can be predicted, preventive action can be taken to mitigate the pending failure. This research work aims at developing a novel framework that applies machine learning and probability theory based on data mining techniques using significant amount of captured IT equipment fault log data from a central server to predict faults. Statistical test based on Naive Bayes and K-Nearest Neighbour classifiers were used and implemented using Rapid miner running on java programming language. The results obtained were models showing good prediction results with accuracy of 83% and 27% respectively indicating substantially, that applying data mining in equipment fault prediction is possible with datasets features that are best fit.

Keywords: Fault, machine learning, data mining, Naive bayes, k nearest neighbour(knn)