Reliable predictions are an invaluable tool in all forms of trading. To be able to predict a future demand in advance with a reasonable margin of error means that ICA can buy regular volumes of goods. In this way, minimal storage is required which means lower costs. A common approach is to use historical marketing and sales data to create linear forecasting models. These have many advantages, eg they are easy to apply and can be used to give an explanation for a particular outcome. Normally they function relatively well in stable conditions, eg when the brand is selling well and not is affected by temporary factors, such as marketing activities implemented by the own company or others. However, the system performs relatively poor when changes in large volume occur caused by infrequent factors. Therefore, there is a need to develop models that can merge information from multiple sources and forecast the outcome for this type of marketing. Based on this, one could meet the demand resulting from the marketing activity of the actual product. Today's technology situation also makes it possible to access data from several different sources (internal as well as external), and use this data to achieve more reliable models. For example, it can involve weather information linked to a campaign in order to predict demand for certain goods (eg ice cream in a sunny and hot weather). A desirable situation is when a store can predict both the demand-based marketing, customer’s regular purchasing behavior (which of course depends on the customer base that the current store has) and other important factors. Making a forecast is therefore a process in which several sources of information are to be merged.
The project has as its main goal to improve techniques for data mining.