A Sales Forecast Method for Products with No Historical Data

Abstract

With the acceleration of demand changes, the life cycle of products has been significantly shortened, and new product releases frequently. Thus, enterprises have an urgent demand for sales forecast of new products with no historical data. This article presents a forecasting approach based on the idea of “commodities with similar characteristics, sales may be similar”. We use similarity measures, consistency checks, and feature random search methods to predict sales in the first N months after launch. To illustrate the performance, we employ the model on a car sales dataset. The results show that compare with a series of machine learning models, the proposed method increased the forecast performance in a relatively short time, and it can be used in small size dataset that expands the scope of use and meets the actual business demands.