A machine learning-based framework for forecasting sales of new products with short life cycles using deep neural networks

Demand forecasting is becoming increasingly important as firms launch new products with short life cycles more frequently. This paper provides a framework based on state-of-the-art techniques that enables to use quantitative methods forecast sales of newly launched, short-lived are similar previous when there limited availability historical data for the product. In addition exploiting using time-series clustering, we perform augmentation generate sufficient and consider two cluster assignment methods. We apply one traditional statistical (ARIMAX) three machine learning deep neural networks (DNNs) – long short-term memory, gated recurrent units, convolutional networks. Using large sets, investigate methods’ comparative performance and, larger set, show clustering generally results in substantially lower errors. Our key empirical finding simple ARIMAX considerably outperforms advanced DNNs, mean absolute errors up 21%–24% lower. However, adding Gaussian white noise our robustness analysis, find ARIMAX’s deteriorates dramatically, whereas considered DNNs display robust performance. provide insights practitioners