The cold start demand forecasting problem refers to the challenge of predicting future demand for a new product or service with little or no historical sales data available. This situation typically arises when a company introduces a new product or service to the market or a new product is launched in a store that is already being sold in other stores, and there is no past data available for training a machine learning model to predict future demand.
In traditional demand forecasting, historical sales data is used to train a machine learning model that can predict future demand. However, in the case of a new product, there is no historical data available. This presents a significant challenge because accurate demand forecasting is critical for making informed decisions about inventory, pricing, and marketing strategies.
This second accelerator of a three-part series on demand forecasting provides the building blocks for cold start modeling workflow on series with limited or no history. This notebook provides a framework to compare several approaches for cold start modeling.
The previous notebook aims to inspect and handle common data and modeling challenges, identifies common pitfalls in real-life time series data, and provides helper functions to scale experimentation with the tools mentioned above and more.
The dataset consists of 50 series (46 SKUs across 22 stores) over a 2 year period with varying series history, typical of a business releasing and removing products over time. The test dataset contains 20 additional series with little or no history which were not present in the training dataset.