Micro-reserving for Workers’ Compensation Claims: a case study from a Kaggle Competition

Reserving for Workers’ Compensation claims using Machine Learning & Natural Language Processing (NLP): A case study from a Kaggle competition

Machine learning advances at an unprecedented pace these days. Companies making use of these new solutions might gain an edge over their competitors, while at the same time laying themselves open to critics, for example, for lack of transparency.

The “Actuarial Loss Prediction” (ALP) competition aimed to further this discussion by creating a reserving challenge that goes beyond conventional methods. The challenge was hosted on Kaggle, the leading platform for data science competitions with a community of over 8 million data scientists, and was presented as follows:

The Actuaries Institute of Australia, Institute and Faculty of Actuaries and the Singapore Actuarial Society are delighted to host the Actuarial loss prediction competition 2020/21 to promote development of data analytics talent especially among actuaries. The challenge is to predict Workers Compensation claims using highly realistic synthetic data. The data is fully synthetic and not specific to any legal jurisdiction or country. We are grateful to Colin Priest for building and supplying the dataset. We invite the competitors to take claims inflation into account.“ For further details, please see https://www.kaggle.com/c/actuarial-loss-estimation.

The ALP ran from December 2020 to April 2021, keeping 140 teams of actuaries and data scientists busy. We have experimented with various model architectures, of which a boosted random forest proved to be the most performant. In the final model information in the claim descriptions was extracted using NLP and the ultimate incurred claim cost was predicted using a maximum-likelihood method. At the end we retained the second place and valuable insights about how machine learning in reserving might look like in the future, which we share in this paper.

Find the Q&A here: Q&A on 'General Insurance Reserving'