Forecast of individual customer’s demand from a large and noisy dataset

Optimization of the supply chain relates on data that describes actual or future situation. Besides in many situations available data may not correspond directly to what is expected for the different models because of too large quantity and imprecision of the data that may lead to suboptimal or even bad decisions. Actual problem considers the availability of a large and noisy dataset concerning historical information about each customer that will be used to make improved prediction models, that may fit models to optimize the supply chain. When dealing with large datasets, market segmentation is frequently employed in business forecasting; many customers are grouped based on some measure of similarity. Segment-level forecasting is then employed to represent the population within each segment. Challenges with successfully applying market segmentation include how to create segments when descriptive customer information is lacking and how to apply the segment-level demand forecasts to individual customers. This research proposes a method to create customer segments based on noisy historical transaction data, create segment-level forecasts, and then apply the forecasts to individual customers. The proposed method utilizes existing data mining and forecasting tools, but applies them in a unique combination that results in a higher level of customer-level forecast accuracy than other traditional methods. The proposed forecasting method has significant management applications in any domain where forecasts are needed for a large population of customers and the only available data is delivery data.