Machine Learning (ML) techniques have been applied broadly in the field of actuarial science and achieve fruitful results. However, to solve a practical problem with ML, the agent needs to preprocess the dataset, choose the proper ML tools, and tweak the hyperparameters of the model. All these steps are task-specific and require expert knowledge, which may render less effective ML model performance when operated by a non-ML-expert agent. To this end, AutoML is developed to facilitate the usage of ML by non-expert agents and aims to integrate the automated data processing, model selection, and model tuning into one system. This project develops two versions (Python and R) of AutoML pipeline that adapts and modifies the existing AutoML packages (for example, caret, AUTO-SKLEARN) to fit better for the actuarial ML problems. We will perform, but not limited to, the following tasks:
- Literature review on the existing AutoML system
- Understand and modify the existing AutoML codebase
- Wrap up the results in a scientific report
Supervisors: Zhiyu (Frank) Quan, Xiaochen Jing
Graduate Supervisor: Yuxuan Li