Interval forecasting with grey models: a novel learning procedure for improved decision-making

Purpose This study addresses two key challenges in interval grey number prediction extensions: information loss in transformed sequences and sensitivity to weight selection. The research proposes a novel modeling approach for short-term time series forecasting, particularly in environments with limited data availability. Design/methodology/approach The proposed approach integrates boxplots and Program Evaluation and Review Technique (PERT) with grey modeling through three main steps: data fuzzification using boxplots to infer upper and lower sequences, model construction employing various forecasting models and prediction aggregation using PERT concepts. Findings The proposed procedure was validated using four datasets: two from previous studies, one from a simulation case and one from the UC Irvine Machine Learning Repository. The experimental results demonstrate that the proposed approach achieves superior prediction accuracy compared to Grey Model and its extensions. Practical implications This approach is particularly valuable for industries with limited data availability. It provides a more reliable method for decision-making in environments where traditional data-driven approaches may be insufficient due to small sample sizes or fragmented datasets. Originality/value The study introduces a novel combination of boxplots and PERT with grey modeling, offering an innovative solution to overcome the limitations of current interval grey number prediction methods. This approach provides a more robust framework for handling uncertain and non-uniform numerical distributions in short-term time series forecasting.