Promo Detection: Time Series Classification methods applied on CPG data

Promotion detection is an important task for businesses as it can help to increase sales and revenue. However, accurately detecting promotional periods can be challenging due to the scarcity of available datasets, noisy labels, and imbalanced classes. In this work, two Deep Neural Network (DNN) models were implemented for the task of promo flagging, with the objective of substituting or acting as a decision support to business analysts. Three datasets of different sizes and labeled by different analysts were used to evaluate the models' performance. The results show that the DNN model can perform better than rule-based models and can primarily rely on raw features such as units sold for predictions. The addition of a trainable model to the flagging workflow can increase productivity and reduce costs. The accurate detection of promotional periods can help predict the effectiveness of future promotional strategies, which is particularly important as promotional information may not always be available or sold by retailers. Overall, this work highlights the potential of DNN models for promo flagging tasks in the retail industry.