A product life cycle curve prediction method via dynamic fusion and inference rectification

Accurate product life cycle (PLC) curve forecasts of new products present a challenge under data sparsity and volatile demand. Due to the limited availability and underutilization of historical data on new products, conventional forecasting methods (e.g., the Bass model and exponential smoothing models) with pre-specified functional relationships are limited in capturing emergent market dynamics and idiosyncratic product characteristics. To address this issue, the PDFSR-RSI framework integrates Prior-Driven Fusion with Structure Refinement (PDFSR) and Rectification with Statistical Inference (RSI) into a two-stage forecasting process PDFSR can effectively aggregate multiple heterogeneous prototypes from analogous products into a prior PLC curve tailored to the specific characteristics of each new product. Within PDFSR, a Sparsity Optimized Post-Processing (SOPP) mechanism mitigates confidence distortion, while a Random Forest (RF) model enhances nonlinear representation capacity in information-poor environments. Subsequently, RSI dynamically rectifies structural biases through a non-parametric statistical mechanism that leverages adjustment signals extracted from early-stage sales data. Empirical validation using data from Babyonlinedress, a fast-fashion e-retailer, demonstrates that the proposed PDFSR-RSI framework corrects structural biases under data scarcity, thereby improving PLC forecasting accuracy and supporting more robust decision-making in volatile environments.