A New Approach to Forecast Intermittent Demand and Stock-Keeping-Unit Level Optimization for Spare Parts Management

The intermittent and lumpy demand of spare parts requires the choice of the right forecasting model among a variety of existing methods. Spare parts have an uneven lifecycle and mean time to failure for each individual item. As a result, they have a varied time of replacement, and consequently, a varied demand. This paper introduces a multi-cost function optimization approach that dynamically selects and adjusts forecasting models tailored to each spare part. The performance comparisons of the various demand forecasting methods led us to a new forecasting model, the Sfiris–Koulouriotis (SK) method, suited for highly lumpy and intermittent demand. A scaled version of the novel Stock-Keeping Unit-oriented Prediction Error Costs metric is also introduced. The composite negative-binomial–Bernoulli probability distribution for the stock control leveraged the replenishment policy. The best safety stock level is calculated for each individual item. Empirical validation in the automotive industry demonstrated that our approach significantly reduces safety stock while maintaining service levels, offering practical benefits for inventory management.