Portfolio - Gourab Banerjee

Gourab Banerjee

Data Scientist

Resume

1

Python Project

2

Machine Learning Projects

1

Deep Learning Project

About Me

Hi, I'm Gourab Banerjee, a final-year B.Tech student in Information Technology at the Government College of Engineering and Leather Technology. I am passionate about transforming data into actionable insights using Machine Learning, Deep Learning, and Computer Vision.

I have developed end-to-end AI/ML solutions across domains such as finance and automobile, involving EDA, feature engineering, model training, data augmentation, model evaluation, and deployment with Streamlit. My projects include credit risk modeling and a car damage detection system using CNNs, transfer learning (ResNet, EfficientNet), hyperparameter tuning with Optuna, and regularization techniques.

I am proficient in Python, SQL, Scikit-learn, PyTorch, Power BI, and XGBoost, with certifications in ML, SQL, and Statistics from Codebasics. Currently, I am seeking opportunities as a Data Scientist or Machine Learning Intern to apply my skills, deliver real-world business solutions, and grow in this dynamic field.

Key Skills

Python

SQL

Power BI

Excel

Scikit-learn

XGBoost

NumPy

Pandas

Matplotlib

Seaborn

Streamlit

Jupyter Notebook

Google Colab

JSON

Git & GitHub

FastAPI

Data Cleaning

Exploratory Data Analysis (EDA)

Feature Engineering

Supervised Learning

Unsupervised Learning

Classification Models

Regression Models

Model Evaluation

Cross-Validation

Hyperparameter Tuning

Model Deployment

Statistical Analysis

Data Visualization

SMOTE

PyTorch

TorchVision

Computer Vision

Convolutional Neural Networks (CNN)

Feed Forward Neural Networks

RNN

Transformers

Transfer Learning (ResNet, EfficientNet)

Data Augmentation

Regularization (Dropout, L2)

Hyperparameter Tuning (Optuna)

My Projects

Car Damage Detection

Domain/Function: Automobile & Computer Vision

Health Insurance Cost Prediction

Domain/Function: Healthcare

Credit Risk Modelling

Domain/Function: Financial and Banking

Customer Churn Prediction

Domain/Function: Telecom

E-commerce Sales Dashboard

Domain/Function: E-Commerce

My Experience

Academic Machine Learning & Deep Learning Projects

Jan 2025 – Present

Built and deployed end-to-end ML/DL models across domains such as finance and automobile, focusing on real-world problem-solving and deployment.
Developed a Credit Risk Model achieving 92.3% accuracy; applied SMOTE for class imbalance, evaluated with ROC-AUC, and deployed using XGBoost and Streamlit.
Designed a Car Damage Detection system using CNNs in PyTorch, applied data augmentation on ~2,300 images, leveraged transfer learning (ResNet, EfficientNet), and optimized with Optuna and regularization, achieving 78% accuracy.
Gained hands-on experience in Python, SQL, Scikit-learn, PyTorch, XGBoost, Power BI, and real-time ML/DL deployment.

Awards & Certificate

Python: Beginner to Advanced For Data Professionals

SQL for Data Science

Math and Statistics For AI, Data Science

Master Machine Learning for Data Science & AI: Beginner to Advanced

Deep Learning: Beginner to Advanced

Let's Connect

Feel free to get in touch with me. I am always open to discussing new projects, creative ideas or opportunities to be part of your visions.

Download Resume

Resume

Mail Me

[email protected]

https://www.linkedin.com/in/gourab-banerjee-8b3a98185/

Call Me

+91 9732385249

GitHub

https://github.com/BanerjeeGourab

Deep Learning Project

Car Damage Detection

Domain/Function: Automobile & Computer Vision

Developed a CNN-based model in PyTorch to classify car images into No Damage, Minor Damage, and Severe Damage. Used data augmentation, transfer learning (ResNet, EfficientNet), Optuna for hyperparameter tuning, and deployed via Streamlit.

View Project
Engagement

View Project
on GitHUB

View Project
on Jupyter Notebook, Streamlit

Other Tools

PyTorch

TorchVision

CNN

Transfer Learning

Streamlit

Deployed Streamlit app for Vehicle Damage Detection where users upload car images to get real-time predictions of damage severity.

Streamlit app predicts vehicle damage class from uploaded image — example shows detection of front breakage.

Project Details

Car Damage Detection – Automobile & Computer Vision
Developed a deep learning-based system using Convolutional Neural Networks (CNNs) in PyTorch to classify car images into No Damage, Minor Damage, and Severe Damage categories. Worked with a limited dataset (~2,300 images) and applied data augmentation to improve generalization. Leveraged transfer learning with ResNet and EfficientNet, achieving the best performance with ResNet (78% accuracy) after applying regularization techniques to reduce overfitting. Conducted hyperparameter tuning with Optuna to optimize model performance. Deployed an interactive Streamlit application that enables users to upload car images (front/rear view) and receive real-time predictions for damage assessment. This project demonstrates expertise in Computer Vision, Deep Learning, Model Optimization, and End-to-End Deployment, with potential applications in the automotive and insurance industries.

Machine Learning Project

Health Insurance Cost Prediction

Domain/Function: Healthcare

Developed a Health Insurance Cost Prediction model using regression algorithms. Performed EDA, feature engineering, and model tuning. Achieved strong performance and deployed a user-friendly Streamlit app for real-time premium estimation based on user attributes.

View Project
Engagement

View Project
on GitHUB

View Project
on Jupyter Notebook, Streamlit

Other Tools

Streamlit

Python

XGBoost

Statistics

Streamlit app to predict health insurance cost using user data like age, income, BMI, medical history, and lifestyle factors.

Plotted histograms with KDE for key numeric features to analyze distributions and trends using Seaborn in a 3x2 grid layout.

Project Details

Built a machine learning web application to predict health insurance premium costs based on individual demographic and health-related attributes. Performed data cleaning, exploratory data analysis (EDA), and feature engineering to identify key cost-driving variables such as age, BMI, smoking status, and medical history. Trained and compared multiple regression models including Linear Regression, Random Forest, and XGBoost, optimizing performance using cross-validation and hyperparameter tuning. Evaluated models with MAE, RMSE, and R² scores, selecting the best-performing model for deployment. Developed a user-friendly interface using Streamlit, allowing users to input personal details and receive real-time premium estimates. Technologies used: Python, Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn, and Streamlit.

Machine Learning Project

Credit Risk Modelling

Domain/Function: Financial and Banking

Developed a Credit Risk Model using logistic regression and XGBoost to predict loan default probability. Achieved 92.3% accuracy. Handled class imbalance with SMOTE, performed EDA, feature engineering, and deployed the model via Streamlit.

View Project
on GitHUB

View Project
on Jupyter Notebook, Streamlit

Other Tools

Python

XGBoost

Statistics

Streamlit

FastAPI

Interactive Streamlit app for predicting credit risk using financial inputs like income, loan details, and delinquency ratios.

Visualized feature correlations using a heatmap to analyze multicollinearity and understand relationships impacting loan default prediction.

Project Details

Built an end-to-end Credit Risk Prediction model to identify potential loan defaulters using historical financial and demographic data. Performed extensive EDA and data cleaning, followed by feature engineering to improve model accuracy. Addressed class imbalance using SMOTE and applied multiple algorithms including Logistic Regression and XGBoost. Achieved 92.3% accuracy and strong ROC-AUC performance. Conducted hyperparameter tuning and model evaluation using confusion matrix, precision, recall, and F1-score. Deployed the final model using Streamlit for real-time risk prediction. The app allows users to input customer data and receive instant predictions with confidence scores. Tools and technologies used include Python, Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn, XGBoost, SMOTE, and Streamlit.

Python Project

Customer Churn Prediction

Domain/Function: Telecom

Performed Exploratory Data Analysis (EDA) on a Customer Churn dataset to identify key factors influencing churn. Analyzed trends in demographics, usage patterns, and service features using visualizations and statistical summaries to derive actionable business insights.

View Project
Engagement

View Project
on GitHUB

Other Tools

Numpy

Pandas

Matplotlib

Seaborn

Displays EDA on customer churn data: structure, null values, and data type conversion for "TotalCharges" in a telecom dataset.

Visualizes churn distribution by gender and senior citizen status using bar plots; highlights churn trends across demographic groups in telecom data.

Project Details

Analyzed a telecom dataset with 10,000+ customer records to understand churn patterns.

Performed data cleaning (handling nulls, correcting types) and preprocessing.

Conducted univariate and bivariate analysis to explore distributions and relationships.

Identified key features influencing churn such as contract type, tenure, monthly charges, and customer service interaction.

Discovered higher churn in month-to-month contracts, low tenure customers, and users without tech support.

Applied correlation heatmaps and count plots to visualize impactful factors.

Generated insights to support churn reduction strategies and improve customer retention.

Tools used: Python, Pandas, Matplotlib, Seaborn, Jupyter Notebook.

E-commerce Sales Dashboard

Domain/Function: E-Commerce

An interactive eCommerce sales dashboard displaying key metrics like total amount, profit, quantity, and AOV. It includes visual insights on profit by state, month, and sub-category, customer performance, payment modes, and product categories with quarterly and filter options.

View Project
Engagement

Other Tools

Power BI

Excel

An interactive eCommerce sales dashboard displaying key metrics like total amount, profit, quantity, and AOV. It includes visual insights on profit by

Project Details

This project involved building an interactive Power BI dashboard to analyze and visualize eCommerce sales data. The dashboard presents key performance indicators such as total sales amount (438K), total profit (37K), quantity sold (5615), and average order value (121K). It offers dynamic visualizations including profit trends by month, state-wise performance, and sub-category profitability. Donut charts depict distribution of payment modes and product categories, while bar charts highlight top-performing customers. The dashboard also includes slicers for quarterly and category-based filtering, enabling users to interactively explore trends and gain actionable business insights. The overall goal was to provide a comprehensive, visually appealing, and user-friendly tool to monitor sales performance and support strategic decision-making. Tools used: Power BI, with a focus on DAX, filters, and formatting best practices.