WA L&I claim auto-adjudication predictive modeling system is a 100% automated system using both predictive modeling models (around 100 machine learning models) and 70 business rules to adjudicate simple, predicted, inexpensive new claims. The daily output tables and over 40 control charts can be accessed by webpage (R Shiny). Since 2013, it has been rewritten four times. The successful system consists of four parts: i) stable data input, ii) predictive models, iii) business rules, and iv) deploying the automated system with flexible output. Some of the significant challenges, e.g., killing and filling the database, bias, and trade-off with uncertainties, modeling ICD10 code, how to fill in missing values, transforming categorical variable to real continuous variable and vice versa, choice of predictive models, modeling closure time, model selection and validation, and suggested solutions are provided with examples. Key takeaways and future developments will be discussed.
Learning Objectives:
1. Understand the pros and cons of automation in building predictive modeling systems.
2. Transform predictor variables from categorical to real continuous, or vice versa.
3. Explore automated auto-adjudicated claim results displayed in R shiny package.
