In this thesis a system for knowledge discovery is developed that ensures interaction of continuous (short time series of demand observations) and discrete (parameters that describe the product) attributes. As a result, a demand forecast is obtained with whose help the process under consideration is managed. The elaborated system can be applied in different areas related to decision making when forecasting a new product, which might reduce the risks associated with new products. Continuous data generalisation related to data collection is performed. Data pre-processing includes various processes, e.g., noise clearance, non-informative data removal, as well as data transformation and normalisation. The interaction of data collection and pre-processing processes is studied. For the data derived in the course of pre-processing, clustering is employed to determine an optimal count of clusters. As a result of clustering, for the objects with historical demand their membership in a class is determined by modifying the k-means algorithm. Based on the clusters found, prototypes are formed that are used for forecasting a new product. Different techniques are examined with a view to improving clustering results. New data classification was made using inductive inference on the basis of decision trees with the help of C4.5 algorithm. The process of classification is analysed and the results obtained are evaluated. Developed system performance check and its evaluation was performed using a separate data set. The obtained results of classification processes and cluster analysis interaction are analysed. Application of forecasting results to the management of the process under consideration is studied. The obtained research results are generalized in tables and graphics for visualization.