Sellings Predictor
All businesses manage their stocks with more or less advanced methods. Many times depends on the manager's experience or intuition. But when the number of references is very large, using machine learning reports a great advantage, since the predictions can be much more closely adjusted to reality, especially in medium-sale products that are not as present in manual controls.
Introduction
Our project on sellings predictor is used by Agapea bookchain to order books automaticaly. Every night Agapea sends to our servers the last day selled on each shop and the system generates the predictions on what books are going to be selled and how many for the next weeks. The predictions are really accurate, and generate authomatic orders to providers. Managers handle manual orders only on special books, like a title needed for an author's book signing
Research
The develpment was done using LSTM networks on Keras with Tensorflow as backend.
The results was very good. Now we are testing PRModel to explore it as an better alternative to LSTM.
Final Result
The impact in the Agapea bussines has been big with 60% less stocks breaks
less stock breaks