Imagine being a shoe retailer in Romania, trying to set the perfect price for your products to maximize profit and sales. This is exactly what I’m working on, using machine learning to optimize shoe pricing for a retailer in Romania. My goal is to predict the optimal price for each shoe size, taking into account various factors like product attributes, pricing and cost info, historical sales, store-level stock levels, daily weather, and calendar info.
I’m using a feedforward neural network (FNN) to predict two targets: profit and quantity sold, both over a 14-day forecast horizon. But here’s the catch: I need to avoid selling too much or too little by a certain date. So, how do I set per-day sales limits or control the pace of sales within this forecast horizon?
My initial thought was to evenly split stock across days and set a cap, but this approach ignores natural daily fluctuations like weekend demand spikes or weather-driven changes in sales. I’m looking for better ways to model daily caps or sales pacing, ideas for incorporating seasonality or constraints directly into training, and alternatives to FNN for this type of structured data.
Do you have any feedback on my modeling strategy or optimization approach? I’d love to hear your thoughts on how to tackle this challenge.