User Tools

Site Tools


8a.013.SBU - Medical Insurance Claim Prediction

Project - Summary

Predicting the risk of certain health conditions people are going to have in the future is very useful for both health insurance providers and customers. This task involves obtaining representation of unstructured and irregular health care data and building precise and efficient prediction models. This project uses purely claim records from health insurance providers to tackle these two problems. Code2Vec is introduced as a code embedding model to learn the embedding vectors for numerous ICD diagnosis codes. From the embedding, each health insurance customer can be represented by a sequence of feature vectors. An LSTM network is trained on these sequences to make predictions of future health insurance claims. The experiments show that the embedding vectors are well-interpretable. And the LSTM model have utilized the temporal information in the health insurance claim records to make precise predictions.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Minh Hoai Nguyen PI (631) 632-8460 Stony Brook University
Zekun Zhang Student Not available Stony Brook University
Vinh Tran Student Not available Not available Stony Brook University
Funded by: Softheon

Project - Novelty of Approach

  • Address an important and novel problem
  • Has strong integration with member company
  • Has broader impacts to other applications
  • Preliminary results and lessons have been obtained

Project - Deliverables

1 A data-mining algorithm for identifying predictable patterns
2 A machine learning algorithm for improving the identification of predictable patterns
3 A dataset and a list of predictable patterns

Project - Benefits to IAB

Health insurance prediction is very useful for both health insurance providers and customers. Insurance providers need this kind of risk assessment to determine the terms and rates of insurance plans. Customers benefit from this by choosing the most suitable plans to save minimize spending. Traditionally, such assessments are carried by professional actuaries and clinicians, using customers’ personal profiles and medical history. There may be some models involved, working on hand-crafted customer features and well-defined risk factors. But these models require efforts from expertise, and usually do not generalize to other situations. The proposed method assumes that there exists relations between customers’ previous claims and future claims. The prediction models utilize this kind of relations to make accurate and autonomous predictions.

Project - Documents

FilenameFilesizeLast modified
8a.013.sbu_final_report.docx578.4 KiB2020/07/30 15:40
8a.013.sbu_year_8_pitch.pptx123.2 KiB2019/09/18 17:28
8a.013.sbu_year_8_project_proposal_minh_hoai_nguyen.pdf609.0 KiB2019/09/18 17:25

Life Form Feedback

Year 8 Project Poster Session Feedback (Fall 2019)
Real name Great Progress On Course Needs Change Off Course Abstain
Kimmo Valtonen (kimmo.valtonen)     
Sumit Shah (sumit.shah)     
projects/year8/8a.013.sbu.txt · Last modified: 2020/07/30 15:55 by sally.johnson