Analytics Case Study
Payment Stream Analysis Algorithm
After applying several well-known clustering algorithms, our team developed a new, proprietary algorithm to meet our client’s needs.
The Problem
Our client wanted to develop an automated process that would analyze both the frequency and amount of current and historical payments made by non-custodial parents, to predict future monthly income. An automated process would also help to ensure compliance with policy guidelines, and reduce the time, errors and inconsistencies resulting from a manual review process.
Our Solution
Our team analyzed a stream of child support payments and classified the data based on payment frequency and amount. After preprocessing and transforming the child support payment data into a more structured format, we applied several well-known clustering algorithms such as Dynamic Time Warping, K-Means, Random Forest, and Support Vector Machine to the data. A detailed analysis of the results found shortcomings with each of these algorithms.
In response, our team developed a new, proprietary algorithm. Manually tagging the payment streams into the appropriate clusters to mimic the “Mechanical Turk” technique, we applied the new algorithm to the payment data and compared the results with those of all the algorithms originally applied. The new algorithm was demonstrated to be the most effective, and was integrated and implemented as an analytics-based classifier model, and applied to payment streams in real time.
Challenges
- Developing a new, custom algorithm that would improve over existing ones, Validating data
- Visualizing and interpreting data,
- Ensuring scalability of the solution
- Adhering to stringent privacy and security requirements
- Complying with a growing array of State and Federal regulations
- Providing complete audit functionality
Benefits
- Immediate, cost-effective solution
- Speedier eligibility determination
- Increased productivity and accuracy
