Aakash Sekar

Expand Transaction Data and Enhance Merchant Tagging

• Implemented the scaling of credit, debit, and private label card transactions through the utilization of advanced machine learning models, aiming to accurately reflect the entire United States population
• Applied NLP techniques at TransUnion to fine-tune machine learning models for processing hundreds of millions of transaction records, resulting in the creation of a comprehensive merchant name mapping table
• Developed and fine-tuned machine learning models using NLP, capable of efficiently processing hundreds of millions of transaction records

Forecast holiday spend using time series model

• Developed and deployed time series models using four years' worth of credit card transaction data to predict holiday spending for individual merchants
• Successfully adapted the model to effectively handle the unprecedentedly low revenue outliers triggered by the COVID-19 pandemic, particularly within industries such as restaurants and airlines
• Executed the implementation of nearly 1500 time series forecasting models, covering the entire country and operating at different granularities, including state, merchants, and industry segments

Evaluation of Ad Campaigns through Randomized Control Trials

• Analyse post ad exposure data to determine the efficiency of any given ad campaign using randomized control trail - a statistical method to evaluate the potency of any medicine
• Worked on control group methods to measure the effectiveness of an ad campaign based on purchase information from credit card transactions
• Developed and optimized machine learning models using multiple features to explain the purchasing pattern seen in our client's transaction data

SKU Modelling for Targeted Advertising Optimization

• Engineered SKU models by leveraging consumer purchase records and Epsilon data to identify the most promising audience for targeted ads across various products
• Collaborated with clients such as Doordash, Whataburger, and Burger King to refine their ad targeting strategies based on the audience insights generated by the model.
• Advanced real-time anomaly detection capabilities through implementation of machine learning-based monitoring systems.
• Enhanced business decision-making with robust statistical analyses, including regression modeling and hypothesis testing.

Fraud Detection in Nationwide Medical Billing Claims Data

• Implemented a fraud detection model utilizing medical billing claims data across the United States, where medical providers are flagged based on predicted model scores
• Addressed numerous business requests, dedicating over 18 months to incorporating new medical conditions and updating the model
• Successfully optimized the End-to-End batch process of the model, achieving a time savings of 3 hours by reducing the processing time from over 15 hours to a more efficient 12 hours.

Fraud detection model to identify suspicious insurance claims using Hotspot analysis

• Implemented a Hotspot analysis for lawyer data within the Personal Auto Insurance domain, leveraging shape files to create a geographical representation of Exploratory Hotspot analysis at the Zip code level
• Implemented deep learning models using TensorFlow, enhancing the accuracy of fraud detection algorithms by 10%
• Devised an anomaly detection model utilizing Exploratory Spatial Data Analysis (ESDA) and Pandas within the Apache Spark framework to statistically identify suspicious insurance claims across the United States
• Customized the Hotspot model to determine the nearest lawyer based on the incident location in cases of legitimate insurance claims.