Mumbai, Maharashtra, India
Link to my Github Profile: t.ly/trwY Self-driven professional with proven experience in managing distinct programs such as carrying out due-diligence on financial credit, assessment of credit risks, and monetization of patented technology by engagement in problem-specific research inquiry and use of analytical techniques. Passionate about exploring the real-life applications of data analytics to address industry problems whether they relate to process developments, business acquisition, or better user experience of services. Reported to people in senior managerial positions in past organisations and was involved in important assignments that allowed me to demonstrate my analytical skills. Snapshot of my analytical experience: Please check my profile on Github to view the Analytics projects completed by me. In addition, the highlights of my experience in analytical projects are the following: 1. Incorporated industry domain knowledge to observe the lack of features important to justify the default likelihood of clients in given data and build a good classification model. 2. Adopted problem-specific F2-score metric to assign more weightage to recall/sensitivity than to precision to evaluate model predictions for classification of bad loans. 3. Addressed imbalanced misclassification costs in classification of bad loans by employing cost-sensitive learning approach constituting SMOTE (Synthetic Minority OversamplingTechnique) and ENN (Edited Nearest Neighbors). 4. Performed basic ETL (Extract, Transform and Load) operations in Python during analysis of the accelerometer sensor data stored in 3200 text files. 5. Segmented FFP (Frequent Flyer Program) customers of an airline by using K-means clustering and assessed optimal number of clusters or K value by elbow plots. Presented problems in increasing airline's FFP enrollments from customer segmentation in given data by root cause analysis,including contributing external factors, insufficient FFPenrollment duration, and absence of airline's rewards criteria details to understand the flying behavior of FFP customers. 6. Deciphered patterns in the data and dependencies between features by using cross tables, correlations, statistical tests, and visualization techniques in Python and Tableau. 7. Aced problem on prediction of insurance amount by using mutual information to check dependencies and permutation-based feature importance in Random Forest Regressor to rank features and identify overfitting of the model. Please check Kaggle: https://www.kaggle.com/aditi86/30daysofml-aditi
Consulted Reliance in CBM Division.
Client - Fortune 500 Pharmaceutical (NBA or Next Best Actions Algorithm Team) 1. Built, deployed, and reviewed internal NBA rule-based recommendations algorithms targeted for medical representatives, and configured parameters through writing of complex SQL queries on client’s Snowflake web interface. 2. Performed anomaly detection by using IsolationForest and did forecasting by using Prophet on past 5-year time-series recommendations IQVIA data segmented by geography and frequency, applied GridSearchCV for hyperparameter tuning, and Matplotlib for visualizing anomalies in Azure Databricks environment. Retrieved, cleaned, and analysed data from parquet files stored in Azure layer using Spark SQL and saved as Delta table in Unity Catalog. Presented findings to the senior management. 3. Did grouped time-series forecasting for quarterly drug prescription volumes using the Diviner library in Databricks, performed cross-validation to avoid overfitting, utilized MLFlow to manage the ML lifecycle, and calculated forecasting errors.
Client - Small Industries Development Bank of India (Risk Management Vertical) 1. Extracted insights by consolidating observations from varied past credit datasets such as loan product MIS, rating models, and customer demographics, and prepared dashboards by using Plotly (Python) to identify key features for modeling the Probability of Default (PD) and Loss Given Default (LGD) across various client profiles. 2. Developed a ranking framework to support industry-specific lending decisions. Designed a relative risk metric to draw quantitative analogy between external credit agency reports and internal industry-wise stress accounts scores. Used complex algorithms such as QuantileTransformer, LogTransformation, etc. and weighted aggregation techniques to address skewness and scale the data. 3. Devised migration matrices to track shifts in risk profiles across different industries and analyzed trends in sectoral risk ratings over recent years.
Client - An Indian government oil and gas organisation 1. Learned Julia within two days to contribute to a seismic data viewer project, then trained team members, and utilized Julia libraries for visualizing 2D and 3D seismic data files (.seis format). 2. Translated Julia code for seismic data analysis into Python, ensuring compatibility with web applications on the client’s Django server, leading to the successful deployment of the application. 3. Developed a stacked ensemble regression model (Python) to predict the Rate of Penetration (ROP) in oil well drilling, optimized drilling parameters using Cat Swarm Optimization model from Mealpy, and benchmarked model performance using metrics like R-squared, RMSE, and MAPE for improved prediction accuracy.