Companies nowadays have to deal with shorter product life cycles, which increases the need to properly forecast demand for new products. Forecasts allow them to make operational decisions, such as procurement and inventory control. The main purpose of the thesis is to forecast sales of new products in the FMCG sector. Firstly, we acquired access to appropriate data and transformed it so that it can be used in different machine learning models. We solved the problem by implementing a method called DemandForest. Besides point forecasts the method can establish prediction intervals. We evaluated DemandForest multiple times with different number of clusters. On the basis of our experimental results, we discovered that DemandForest provides more accurate results than the benchmark method.
|
