Abstract
{ "background": "The management of industrial machinery fleets represents a significant capital and operational expenditure for engineering firms. In South Africa, the selection and evaluation of fleet management methodologies have historically relied on deterministic cost models, which inadequately capture the hierarchical and temporal variability inherent in operational data.", "purpose and objectives": "This study conducts a comparative methodological evaluation to determine the most robust analytical framework for measuring the long-term cost-effectiveness of different fleet management systems. It aims to quantify the predictive performance of multilevel regression against traditional pooled and fixed-effects models.", "methodology": "A longitudinal dataset comprising maintenance records, utilisation metrics, and total cost of ownership for four predominant fleet systems was analysed. A three-level linear mixed model was specified: $y{ijt} = \\beta0 + \\beta X{ijt} + u{i} + v{ij} + \\epsilon{ijt}$, where $ui$ and $v{ij}$ are random intercepts for fleet type and individual asset, respectively. Model comparison was based on the Watanabe-Akaike information criterion (WAIC) and out-of-sample predictive accuracy.", "findings": "The multilevel model demonstrated superior predictive performance, reducing root mean square error in cost projections by approximately 18% compared to the best-performing traditional model. Crucially, the 95% credible intervals for the random effects revealed that fleet-type variance accounted for nearly 35% of the total unexplained variance in annual costs, a component missed by pooled models.", "conclusion": "Multilevel regression provides a statistically rigorous and more accurate methodological framework for comparative cost-effectiveness analysis of machinery fleets, as it explicitly models the clustered structure of the data.", "recommendations": "Engineering managers and analysts should adopt hierarchical modelling techniques for fleet investment appraisals. Further research should integrate real-time sensor data into the model's predictive layers.", "key words": "fleet management, cost-effectiveness, multilevel modelling, regression analysis, industrial machinery, predictive maintenance", "cont