A Hybrid Forecasting System for New Product Demand: Integrating Statistical and Machine Learning Methods

Forecasting demand for new products remains a major theoretical and practical challenge because forecasts must be generated when little or no product-specific sales history is available. This study proposes a hybrid forecasting system that integrates complementary theoretical perspectives from time-series analysis, statistical learning, machine learning, clustering, classification, regression, forecast combination, and uncertainty quantification.

The methodology decomposes new product demand into two components: temporal shape and total demand magnitude. Historical products are characterised through time-series features and grouped into homogeneous demand profiles using clustering techniques. Cluster centroids represent typical demand trajectories and provide the statistical basis for estimating the expected shape of a new product’s demand. Supervised classification models subsequently assign the new product to the most appropriate profile according to its commercial, categorical, and operational attributes.

In parallel, regression-based machine learning models estimate the expected total demand volume. The final forecast is obtained by combining the predicted temporal profile with the estimated magnitude, allowing different statistical and machine learning models to contribute according to their respective strengths. The framework also incorporates calibration, prediction intervals, explainability, and pseudo-new-product back testing to assess forecast accuracy and robustness.

Beyond its methodological contribution, the proposed system is designed for implementation in real organisational environments. It supports forecasting at the pre-launch or early-launch stage and can incorporate product information commonly available to businesses, including category, packaging, price, target market, and distribution channel. Its modular and scalable structure enables integration into existing analytical processes, supporting production planning, inventory allocation, supply-chain coordination, commercial decision-making, and risk reduction during new product introductions.