MAHSA GHANBARPOUR MAMAGHANI

Workers compensation fraud detection model:

• Developed Text-mining technique to read adjuster notes about claims and used a Convolutional Neural Network (CNN) model to predict fraudulent claims
• Developed LGBM model, which combines structure and CNN model results to predict fraudulent claims. The developed model lifted current fraud detection system accuracy by 50 percent based on 6 months of scoring and investigating claims.

Dynamic Pricing Optimization:

• Developed Optimization model to assist setting rate and retention goals for different accounts to maximize expected profitability for various profit centers.
• Developed sensitive analyses method to investigate the trade off between profitability, rate and retention.

Risk Selection Models for XS Casualty:

• Developed risk models, integrated external datasets, and engineered Databricks, MLOps pipeline for unified deployment.
• Automated data ingestion and scoring for multiple business lines.

USC Summarization Pipeline:

• Developed LLM-based NLP pipeline for automated summarization of unstructured claims data.
• Collaborated with cross-functional teams (claims, underwriting) for iterative model enhancement and deployment.
• Applied model validation and experimentation for >92% accuracy and 81% workflow improvement.

Large Loss Prediction Model:

• Built end-to-end supervised learning pipeline (NLP, CNN, XGBoost) to predict large loss claims (>$1M).
• Engineered features from structured and unstructured data; deployed automated dashboards and model monitoring.
• Achieved 20ximprovement in business process accuracy using explainable modeling and data visualization.

Auto Ultimate and GL Ultimate Modeling:

• Led advanced EDA, feature engineering, and model development for predictive analytics and business forecasting.
• Built, validated, and deployed CNN models for ultimate loss projections; automated IBNR calculations.
• Delivered robust model documentation and model monitoring for production workflows.

Proactive surgery cancellation planning due to snowstorms

• Analyzing effects of storms on hospitals' operations -In collaboration with Mass General Hospital using Google BigQuery for collecting and managing patients claims data set (over 650M claims)
• Applied machine learning (GLM, XGBoost, Random Forest), hypothesis testing, and time series forecasting for hospital admissions, discharges, and length of stay under severe weather events.
• Developed forecast evolution scenarios using historical weather forecast data
• Developed reinforcement learning approach to provide the best strategy for hospitals in terms of canceling elective surgeries and appointments based on weather forecasts.
• Developed an implementable heuristic algorithm to provide an alternative strategy for hospitals in terms of canceling elective surgeries and appointments based on weather forecasts

Matching medical staff to long term care facilities to respond to COVID-19 outbreak

• Developed a matching algorithm that pairs healthcare workers with open positions at long-term care facilities

• Developed unsupervised clustering and Elastic Net models for portfolio segmentation and risk assessment.
• Developed an interactive dashboard to demonstrate clusters and correlates of policies' performances

• Converted underwriters assignment problem to a mixed-integer programming model to maximize the total number of submissions.
• Clustered the US to the geographical areas to optimize the workload balance between managers and increased the expected yearly market share for the company

A Novel Multi-Objective Model for Operating Room Scheduling Problem under Uncertainty

• Developed a novel multi-objective model for operating room scheduling problem and assigning patients to hospital units including PACU, ICU and ward
• Developed a Robust optimization model to consider release and operations time uncertainties
• Applied Harmony Search and Simulated Annealing algorithms to solve the operating room scheduling problem