sushma-prog / Advanced-ML

A 7-day hands-on journey into advanced ML techniques including XGBoost, LightGBM, SHAP, LIME, GridSearchCV, and API integration.

📘 Advanced Machine Learning — End-to-End Learning Pipeline

Focus: Real-world pipelines · Model tuning · SHAP & LIME explainability · API integration · Capstone project deployment

---

✅ Week Overview

This week focused on advanced machine learning concepts and practices that go beyond just training models — diving into tuning, interpretation, automation, and deployment-level workflow design.

---

📅 Day-by-Day Breakdown

🔹 Day 1: Introduction to Gradient Boosting

• Revisited bagging vs boosting principles
• Explored XGBoost and LightGBM internals
• Installed and configured libraries
• ✅ Practiced: Trained a basic XGBoost model on the Iris dataset

---

🔹 Day 2: Cross-Validation & GridSearchCV

• Understood cross-validation (CV) strategies: KFold, StratifiedKFold
• Implemented GridSearchCV for hyperparameter tuning
• ✅ Practiced: Tuned max_depth, learning_rate, and n_estimators for XGBoost

---

🔹 Day 3: LightGBM & RandomizedSearchCV

• Compared LightGBM vs XGBoost
• Trained a LightGBM model using Breast Cancer dataset
• Used RandomizedSearchCV for efficient tuning
• ✅ Practiced: Evaluated and compared both models

---

🔹 Day 4: Model Interpretation with SHAP

• Learned why interpretability matters in ML
• Used SHAP to visualize:
  • Global feature importance (summary_plot)
  • Individual prediction reasons (force_plot)
  • Feature interactions (dependence_plot)
• ✅ Practiced: Visual explanations on Breast Cancer dataset

---

🔹 Day 5: LIME + API Basics

• Learned how LIME provides local explanation for black-box models
• Compared SHAP vs LIME
• Used LimeTabularExplainer for per-instance interpretation
• Explored REST APIs: What they are and how to call them with Python
• ✅ Practiced: Parsed real-time public API like cat facts using requests and json

---

🔹 Day 6: Building a Mini Data Pipeline

• Designed an ML data pipeline: Ingest → Clean → Model → Export
• Used:
  • pandas for preprocessing
  • xgboost for modeling
  • joblib for saving models
  • SQLite & CSV for output storage
• ✅ Practiced: Full pipeline built using Titanic dataset

---

🔹 Day 7: Capstone Project — Customer Churn Prediction

• End-to-end ML project using Telco Customer Churn dataset
• Steps performed:
  1. Load and explore dataset
  2. Clean and encode features
  3. Train-test split + XGBoost with GridSearchCV
  4. Interpret model using SHAP & LIME
  5. Save predictions to CSV and SQLite
• ✅ Outcome: A complete, interpretable, production-ready ML pipeline

---

🎯 Tools & Libraries Used

Category	Libraries
Modeling	xgboost, lightgbm
Tuning	sklearn.model_selection
Interpretation	shap, lime
Data handling	pandas, numpy
Deployment Ready	joblib, sqlite3, csv
API Integration	requests, json

---

📌 Key Skills Mastered

• Gradient Boosting Algorithms
• Model Hyperparameter Tuning
• SHAP & LIME Interpretation
• Data Pipeline Design
• API Calls with Python
• Model & Output Storage (CSV, DB)

---

📁 Project Repositories

✅ All completed notebooks and projects from this Week are uploaded here:
🔗 [Advanced-ML](https://github.com/sushma-prog/customer-churn-prediction

---

👩‍💻 Author

Sushma Sandanshiv
BTech Data Science | Aspiring Data Scientist 🔗 LinkedIn • 🔗 GitHub

---