OHAI: Careful Home Care Planning with AI

Important results from the project

The goal of the project was to use machine learning to improve planning by creating a demonstrator of a system that learns from real data from coordinators and home service employees. The project has developed a solution for planning based on reinforcement learning (RL), which is a variant of machine learning.

Expected long term effects

The implemented solution uses real, anonymized data from day-to-day planning; # employees who work on a particular day, their working hours and when they have lunch # clients who have home visits and what home visits to make # which characteristics and competencies to match between employee and client. # Travel times and different modes of travel The result of the AI-run is a proposed planning according to a reward table where different parameters indicate what is defined as "good planning". Planning consists of: choose transport modes, place lunch breaks and home visits.

Approach and implementation

Since the number of selectable combinations to produce a whole planning for all employees was unreasonably many, we chose a strategy that breaks down the problem. The AI modeled each employee as an agent with their own QL algorithm to learn how to plan for their own employees. The agents trained on real data a number of iterations where they, with a certain probability, randomly make a choice and otherwise make the choice with the greatest expected reward. One lesson in the project was to use real data early on since dummy-data easily conceals important problems in the model.