Triveni Kolla

Building Trustworthy AI in Healthcare Finance: An Explainable Machine Learning Framework for Detecting High-Risk Insurance Claims

Abstract:

Healthcare insurers process billions of claims annually, yet traditional fraud detection systems still rely heavily on rigid rule-based approaches that generate high false-positive rates and lengthy investigation cycles. At the same time, advanced machine learning models while achieving superior predictive performance, often operate as opaque “black boxes,” limiting regulatory compliance, investigator trust, and auditability.
This talk presents a production-oriented Explainable AI (XAI) framework designed to classify high-risk healthcare claims while maintaining operational transparency. The architecture integrates ensemble machine learning models including gradient boosting, random forests, and neural networks trained on more than 200 engineered risk indicators derived from claims data such as procedure codes, diagnosis patterns, provider characteristics, temporal billing trends, and cost deviations.
To address the interpretability gap, the framework incorporates multi-method explanation layers, combining SHAP feature attribution, LIME local explanations, rule extraction, prototype-based reasoning, and counterfactual analysis. These techniques produce human-readable explanations for each risk score, enabling investigators, compliance teams, and data scientists to interpret model decisions in real time.
Evaluation against traditional rule-based and baseline machine learning systems demonstrates superior discrimination performance (higher AUC-ROC), increased precision at operational thresholds, and substantial reductions in false-positive claim flags, significantly lowering investigation workloads and payment delays. The framework also improves investigation throughput and improper payment recovery rates, while enabling fairness monitoring across geographic regions, provider specialities, and patient demographics.
Beyond fraud detection, this work illustrates how explainable AI can reconcile predictive power with regulatory accountability in high-stakes financial decision systems. Attendees will gain practical insights into designing interpretable AI pipelines, deploying explainability tools at scale, and building trustworthy machine learning systems for healthcare finance and other regulated industries.  

Profile:

Triveni Kolla is a Senior Business Intelligence Developer with over nine years of experience in designing and developing data analytics and reporting solutions using tools such as MicroStrategy, Tableau, and Power BI. She holds a Master’s degree in Information Systems Management from Marist College, New York, and a Bachelor’s degree in Electronics and Communication Engineering from KL University, India.

Throughout her career, Triveni has worked on developing dashboards, reports, and data visualizations that support business decision-making across various departments, including marketing, finance, and leadership teams. Her technical expertise includes data modeling, SQL development, ETL processes, and building business intelligence solutions that help organizations interpret complex data more effectively.

At Cotiviti, Triveni contributed to business intelligence initiatives that involved developing and maintaining dashboards, improving reporting workflows, and supporting data-driven decision-making. She has experience working with stakeholders and cross-functional teams to understand reporting needs and translate them into effective analytics solutions.

Triveni is also certified as a MicroStrategy Certified Master Analyst and Tableau Data Analyst, reflecting her technical proficiency in business intelligence tools and analytics platforms. She is known for her analytical approach, attention to detail, and ability to collaborate with teams to improve reporting processes and data accessibility.

She remains committed to using business intelligence technologies to transform data into meaningful insights that support informed business strategies and organisational growth.