Adtrac’s retail media customers require precise predictions of customer flows to efficiently sell campaigns to advertisers. This is particularly difficult around shifting holiday and bank holiday dates, which is why we came up with a novel approach to tackle this problem.
Our approach focuses on encoding calendar dates in such a way that a Neural Network then easily can digest this information. This can be achieved by measuring the position of a particular date between its surrounding closed days like public holidays. Moreover, it’s even possible to extend this mapping by taking the number of consecutive closed days into account. In this way, the Neural Network can recognize certain holiday patterns, such as those that occur at Easter, and adjust predictions accordingly. This then helps us to plan campaigns more accurately.
For more details, see our blog post on Medium: https://medium.com/@raphael.schoenenberger_95380/encoding-temporal-features-part-1-f26d08feebd8