Leveraging Feature Ranking for System Fault Identification and Classification Using Machine Learning Algorithms

Abstract Computing environments necessitate the prerequisite of ensuring perfor
mance pinnacle and entailing resilience during disruptions. Most often, system faults
can interrupt the overall productivity and prolong operational time, thereby increasing
computational complexities, along with leading to downtime, mitigated productivity,
and detriments in terms of finances. The inevitability to swiftly effectuate precise
system failure detection, stratification and further resolution, becomes crucial for
effectively maintaining system veracity and evade unethical injections. This indaga
tion pivots on analyzing the various attributes relevant to system fault processing, and
to entail the data thresholding combined with feature extraction to efficiently identify
and classify system failures using machine learning algorithms. The proposed study
entails a multi-modal real-time feature evaluation from the database constructed
using the primary attributes such as the upstream connection, response time, API
latency, connection time, and transaction status. Feature ranking using variance,
Region of Curve (ROC) and T-Test are incorporated to enhance accuracy of classifi
cation. The machine learning algorithms used in this research paper are the Efficient
Linear Support Vector Machine (ELSVM), Naïve Bayes algorithm and Tri-layered
Neural Network, and the performance accuracy rendered by each of the algorithms
are scrutinized. The simulation results are carried out in MATLAB, and the results
are procured successfully.
Keywords System fault identification and classification · Feature ranking · API
latency · Efficient linear SVM · Naïve Bayes · Tri-layered neural network ·
MATLAB