ML based regression for planning of resources in pre-schools and daycare centers

Important results from the project

The goal of this project was to use AI based regression algorithms to improve the prediction of the number of children present at a certain preschool group at a certain time, at say 7, 14 or 28 days into the future. This kind of prediction enables better planning of staffing resources in preschools, as well as e.g. the planning of food distribution.

Expected long term effects

The project was successful. The project resulted in an algorithm that was an improvement, i.e. an algorithm significantly better at predicting the child presence at a preschool department, compared to the current schedule based method. The predictions were better for both short term predictions, i.e a few days into the future, as well as long term predictions, i.e. a few weeks to months into the future.

Approach and implementation

We started the project by doing an analysis of the current data, including examining how well that schedule data reflected reality. We then started using a deep neural network to make predictions on the data. After quite some iterations tests, we came to the conclusion that a regression-based ML model could make pretty much equally good predictions, but it was considerably much faster to train so we switched to this approach instead and this is the model that will be used in production.