Learning optimal policies from event logs through reinforcement learning: a comparison of deep and MDP-based approaches

The paper presents a novel approach to learning optimal behavioral policies from historical event logs using Reinforcement Learning (RL) techniques, specifically a model-based Markov Decision Process (MDP) and a model-free offline Deep RL method. Both approaches aim to optimize Key Performance Indicators (KPIs) in prescriptive process monitoring without extensive domain knowledge. The results indicate that while both methods are effective in improving KPIs, the model-based approach demonstrates superior computational efficiency, which is crucial for practitioners seeking scalable solutions in complex process environments.