Senior Data Scientist
Currentโข ๐ ๐ข๐ง๐๐ง๐๐ ๐๐ง๐๐ฎ๐ฌ๐ญ๐ซ๐ฒ - ๐๐๐ญ๐ ๐๐๐ข๐๐ง๐ญ๐ข๐ฌ๐ญโข Built XGBoost model with 5000+ features to predict fraud detection using iterative approach for training and tuned hyperparameters with Optuna to keep track of best models and hyperparameters โข Analyzed financial data of 40+ key tables with ~ 4 billion records using HiveSQL and PySpark to decide modelling timeframe and feature engineering for modelling over frequency, recency and monetaryโข Generated good/bad transactions v/s summaries for different segments to decide model segmentation criteria using reusable functions with PySpark saving 27% runtime.โข ๐๐ง๐ฅ๐ข๐ง๐ ๐๐ซ๐จ๐๐๐ซ๐ฒ ๐๐ก๐๐ข๐ง, ๐๐ง๐๐ฎ๐ฌ๐ญ๐ซ๐ฒ - ๐๐๐ญ๐ ๐๐๐ข๐๐ง๐ญ๐ข๐ฌ๐ญโข Built Logistic Regression response models with 850 features to predict customerโs response to coupons having 220 liftโข Built Linear Regression revenue models with 220 features to predict revenue generated having 187 lift โข Utilized recursive feature elimination and lasso selection techniques for feature selection to improve lift of models โข ๐๐๐ญ๐๐ข๐ฅ ๐๐ก๐๐ซ๐ฆ๐๐๐ฒ, ๐๐ง๐๐ฎ๐ฌ๐ญ๐ซ๐ฒ - ๐๐๐ญ๐ ๐๐๐ข๐๐ง๐ญ๐ข๐ฌ๐ญโข Built Random Forest and Decision Tree classifier models with 27 features to predict patient attrition with 87% accuracy โข Designed several usecases based on retention, acquisition and profitability and pulled data of about ~5 billion records from 10 sources to size number of patients and scriptsโข ๐๐๐ ๐๐๐ง๐ญ๐๐ซ ๐๐ ๐๐ฑ๐๐๐ฅ๐ฅ๐๐ง๐๐โข Developed resume parser using Spacy, nltk, StanfordNERTagger to parse information like name, education, email, contact with 77% Accuracy.โข Researched and presented complex NLP algorithms like BERT, GPT-2,GPT-3, etc.โข ๐๐๐ฉ๐ฅ๐จ๐ฒ๐ฆ๐๐ง๐ญ ๐๐๐ง๐ญ๐๐ซ ๐จ๐ ๐๐ฑ๐๐๐ฅ๐ฅ๐๐ง๐๐โข Deployed word2Vec model for retail grocery chain using FastAPI, Docker, AWS ECR and EC2 to recommend products to customers.
