Underwriting score prediction for better risk assessment of Autistic cases �
Rajesh Sharma
Contents
What is ASD?
ASD Prevalence among Children & Youth in Canada
ASD Prevalence in Canada
Below graphs are representing:
Asia Region ASD Progression
ASD ASIA Region Progression
Below bar charts are representing:
ASD Health Insurance Expenditure
Data based on research done for some US States
ASD Health Care Expenditure
Below graphs are representing the Total Healthcare Expenditure for “Autism Spectrum Disorder”, “Autistic cases” and “other ASD disorders”.
Rise(%) in ASD Healthcare Expenditure
Below graphs are representing the percentage rise in Total Healthcare Expenditure for “Autism Spectrum Disorder”, “Autistic cases” and “other ASD disorders”.
ASD Diagnosis
ASD Treatment & associated risks
Cost of treatment
Recovery time
Treatment therapies
Problem Statement
How the process currently works?
The only way to predict the risk is using some behavioral questionnaires and ASD leading factors. As, there is no option to identify ASD via medical procedures thus it is hard to assess the risk of the disease.
There 2 core parts of this problem:
Proposed Business Solution
What we are proposing in our solution?
Based on 10 questionnaires and other supporting variables we have build the Underwriting Score ML Model to predict a probabilistic score which will help SunLife in approving or rejecting the policy against ASD cases in the children, adolescents and adults in more accurate manner
Prototype Solution
To build the ML model used the ASD Dataset for Children, Adolescents and Adults available on UCI ML Repository:
Combined dataset contains 10 questionnaires variables and 9 supporting variables with a total of 1100 cases from various countries and ethnicities, and age group.
Solution Design
Prototype Results
Using AWS SageMaker as a platform, Python 3 and its prominent ML libraries created a ML Classifier Model to predict the existence of ASD in a person and achieved a ROC Score of 0.96.
Out of 1100 cases, used 715 cases to train the model and 385 cases to test or validate the model.
Also, predicts the probabilistic underwriting score that is useful to approve/reject the policy or increase the premium amount.
[0-0.3) 🡪 Low risk, [0.3-0.5) 🡪 Intermediate risk, [0.5 >] 🡪 High or Severe risk
Appendix
Top ASD prevalent countries & ASD v/s Age
Worldwide, 1 in every 160 children has an Autism Spectrum Disorder, with 3 times as prevalent among boys as among girls. However, some other well-controlled studies have reported the figures that are substantially higher.
ASD Signs & Symptoms
People with ASD often have problems with social, emotional, and communication skills.
Children or adults with ASD might:
Potential extension for future
Can we predict ASD during pregnancy?
By reading this question you might wonder, how it is related to Health Insurance business or predicting the underwriting score, however, it has a direct link with our ML Model to predict the insurance risk. And, answers to this question can be the potential features for our ML Model.
Now, to predict ASD during pregnancy we actually have to understand how fetal brain develops:
Abbreviations
References
Problem Background
In Canada, among children and youth from 5 to 17 years of age:
An estimated 1 in 66 have been diagnosed with Autism Spectrum Disorder(ASD)
Males were identified with ASD 4x more frequently than females
Solution
Using AWS SageMaker as a platform, Python3 and its other libraries built the ML model on the ASD Dataset for Children, Adolescents and Adults available on UCI ML Repository:
Combined dataset contains 10 questionnaires variables and 9 supporting variables with a total of 1100 cases from various countries and ethnicities, and age group.
Underwriting Score prediction for ASD
Business Impact
Problem Statement
As per the recent strategies of The Province of Canada, Ontario families with an autistic child receives $20,000 a year until their child turns six. After the age of six, families receives $5,000 a year until their child turns 18.
And, it is estimated that Ontario could add $952 million of insurance for autism services including Applied Behavioral Analysis (ABA) and speech/language therapy.
Solution Design