New Product Multimodal Demand Forecasting For Fashion Retail

Abstract

Fashion Industry launches substantial new products every season. Hence, accurate new product demand forecasting is vital for effective demand planning. To tackle this, we propose novel attention based encoder-decoder models that can effectively capture the non-linear relations between product images, attributes, sales and external regressors for robust new product forecasting. Through empirical validation on a large fashion dataset, we show the efficacy and interpretability of our methods as compared to existing baselines.