Machine Learning: Predicting Risks in Insurance
Recommendation algorithms, event predictions, and risk assessments are trending solutions in banks, insurance companies, and many other industries. This is a unique opportunity for companies to optimize expenses, increase workflow speed, and improve service quality.
In this case, we will discuss an IT solution for medical insurance created using machine learning algorithms.
Task
We participated in the development of a solution for insurance companies that automates the patient management process and helps manage the budget.
Solution
Using artificial intelligence algorithms, the system predicts medical episodes, which are all patient visits to a doctor for problems of a specific nature, and can anticipate upcoming episodes. The episode prediction mechanism is built on the analysis of historical data, including information about previously provided services, socio-demographic data (gender, age), and past diagnoses.
The system is complemented by other machine learning models, including:
- detection of uninsured cases with the same diagnosis;
- data on post-discharge patient support methods to prevent relapses;
- recommendations for procedures covered by insurance within 2 weeks after discharge;
- forecasting how the patient will feel after 30, 60, and 90 days.
Their combined use makes it possible to optimize patient management in terms of quality and cost of service.
Result
- A platform was developed from scratch to support the created models.
- The system was trained on a database obtained from patient visits over a span of 3 years using Big Data.
- Integrations with multiple medical systems were implemented.
- Development, testing, and consistent training of ML models were continued.
Technologies
Java, Python, TensorFlow, Amazon SQS, SNS , DynamoDB, Apache Spark