Abhishek Roy

Currently working with a major pharmaceutical client, helping the commercial analytics team, with ML & DS solutions to answer business problems. Leading a team of Data Scientists and engaging with the Client in designing DS solutions for revamping classical business problems and also working on defining and solving new problems which add impact ot the business

1. Market Mix Models

• Build Promotion Response models for zero inflated distributions with segments for responsiveness
• Build Hierarchical Market Mix model based on random effect Bayesian model. Derived Impactable Sales and ROI from these multiplicative model.
• Build Market Mix Optimization solution based on segment level response curves and marginal ROI optimization

2. Persistency & Adherence Analysis

• Created HTML Report from rmarkdown, to capture Persistency Curves of Drugs
• Created patient-drug level data from raw claims data, to capture persistency period and create KM Survival Curve
• Fit custom survival models for skewed distribution and capture censored information to forecast and generate mean persistency
• Calculated patient adherence and identified drivers based on interpretable ML algorithms for customer segmentation

Worked across multiple clients as well as internal learning initiatives. Guided junior data scientists and also worked along them to deliver ML solutions for multiple domains of data science, like:

1.Building automated revenue forecasting pipeline, based on ensemble of multiple time series models

2.Using Survival Analysis and other supervised statistical models to understand the effectiveness of a patient assistance program

3.Using ML Models for Revenue Management for an Airlines Industry Client, by optimizing spillage, spoilage & overbooking

4.Automated Anomaly Detection pipeline as a part of a data management solution focusing on different type of anomalies and variety of data sources

Worked independently, as well as in teams of Data Scientists, across multiple Projects. Also participated in internal Hackathons and other Learning initiatives. Worked across multiple clients as well as some internal research projects like:

1. Clinical Trial Analysis

• Identifying and calculating the Key Success Drivers of a Clinical Trial using Various Supervised ML Models
• Poisson-Gamma Distribution Model for Clinical Trial Enrollment Prediction

2. Market Mix Modeling

• Building a LSTM Model (Keras + Theano) for prediction of Sales using in Panel Data Structure , to capture mixed effect at the panel level.
• Used Explainable AI Algorithms ( LIME, SHAP, MoB) to convert

3. Patient Identification

• Used Ensemble Tree Models (RF & XGB) for Cancer Patient Metastasis Prediction, based on patient claims data across prescriptions, procedures and diagnosis

Created an end to end tool for market mix modelling using Panel Regression model (plm).

• Used Genetic Algorithm in the automated pipeline to get the best model across multiple combinations of feature transformations including promotion adstocks, based on some error metrics and diagnostic tests.
• The backend codes were written on R and the front end was an Excel Dashboard connect via R-Excel

Worked under the guidance of Prof. Bishwabrata Pradhan (ISI), applying copula theory for the estimation of Quality Adjusted Lifetime. All simulation studies and estimation were from real life data and done on R.

Customer Churn Prediction on Telecom usage data

• Created and tested multiple classification models like logistic regression, decision trees , neural network, ensemble models, etc.,
• Compared their performance in terms of accuracy and robustness

Worked under the guidance of Prof. Debasis Sengupta (ISI), focusing on forecasting the monthly repair counts of an electronic item from its historical sales and repairs data. Used survival analysis to predict repair probability over time, and time series forecasting to predict sales