Cost predictions with machine learning
Data analysis using econometric and statistical learning methods
Factsheet
- Lead school School of Health Professions
- Institute(s) Institute of Health Economics and Health Policy
- Funding organisation Others
- Duration 15.03.2020 - 31.03.2021
- Project management Mark Pletscher
- Head of project Mark Pletscher
-
Project staff
Prof. Dr. Tobias Benjamin Müller
Niklaus Meier - Partner Suva
Situation
Project to forecast and decompose medical costs for SUVA accident insurance.
Course of action
Model competition between classic parametric models for cost prediction and machine learning approaches. The machine learning models were each trained in training data and evaluated using test data.
Result
Machine learning approaches (random forests, neural networks, LASSO, etc.) outperform standard approaches in terms of prediction accuracy (in- and out-of-sample). Random forests in particular have emerged as a promising alternative to standard approaches. The decomposition of costs has shown above-average cost growth for outpatient services.
