| The importance of sales forecasting is undeniable. Predicting the sales of the businesses’ products have impact in more than one department of a company. In most cases successful forecasting is a complicated issue especially when the product has not or has just been released to the market hence there are no historical data of sales of that exact product. The current research focuses on addressing a problem that bibliographically is not widely researched, that is forecasting the sales of new fashion products before their market release via an-alyzing their fundamental features and the historical sales data of other, previ-ously released products. To generate accurate results and present a complete strategy various Machine Learning algorithms are modeled, trained, and com-pared to solve the above mentioned problem. The algorithms examined are cat-egorized as non-linear, ensemble and neural networks methods, and the hyperparameters of non-linear and ensemble algorithms are optimized via Grid Search and the hyperparameters of neural networks are optimized via Bayesian optimization. The results reveal that the Convolutional Neural Network (CNN) method is outperforming all the examined algorithms according to Weighted Absolute Percentage Error (WAPE) and Mean Absolute Error (MAE) metrics. No specific category of methods among non-linear, ensemble and neural net-works, was found to perform better. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.