🤖 Machine Learning Projects

A collection of four hands-on machine learning projects, each built as a standalone Streamlit application. Projects are ordered by difficulty and designed to demonstrate key ML concepts with well-commented, learner-friendly code.

Quick Start

# Install all dependencies
pip install -r requirements.txt

# Run any project
streamlit run 01_diabetes_prediction/app.py
streamlit run 02_netflix_clustering/app.py
streamlit run 03_spotify_popularity/app.py
streamlit run 04_churn_predictor/app.py

Each app generates synthetic data on startup so you can run them immediately. See individual project READMEs for instructions on using real Kaggle datasets.

Projects

#	Project	Difficulty	Key Concepts
1	🩺 Diabetes Prediction	2/10	Feature scaling, model comparison, GridSearchCV, ROC-AUC
2	🎬 Netflix Clustering	3/10	KMeans, PCA, elbow method, silhouette score
3	🎵 Spotify Popularity	5/10	XGBoost, LightGBM, SHAP explainability, correlation analysis
4	📊 Churn Predictor	5/10	SMOTE, threshold tuning, precision-recall tradeoff, imbalanced data

Tech Stack

• Language: Python 3.10+
• ML Libraries: scikit-learn, XGBoost, LightGBM, imbalanced-learn
• Explainability: SHAP
• Visualization: Matplotlib, Seaborn
• UI: Streamlit

Project Details

1. Diabetes Prediction with Model Comparison (Difficulty: 2/10)

Use the Pima Indians Diabetes dataset to build and compare multiple classification models. Demonstrates preprocessing, feature scaling, and hyperparameter tuning with visual evaluation.

What you learn: StandardScaler vs MinMaxScaler · Logistic Regression vs SVM vs Random Forest vs Gradient Boosting · GridSearchCV · Confusion matrix, ROC-AUC, F1 score

2. Netflix Show Clustering (Difficulty: 3/10)

Group similar Netflix shows using unsupervised learning based on genre, rating, and duration. Visualize clusters to find patterns in the catalogue.

What you learn: Categorical encoding · Feature scaling for KMeans · Elbow method & silhouette score · PCA dimensionality reduction

3. Spotify Song Popularity Predictor (Difficulty: 5/10)

Predict track popularity from audio features using gradient boosting models. Explain predictions with SHAP values.

What you learn: Feature correlation & multicollinearity · XGBoost/LightGBM tuning · SHAP explainability · Regression model comparison

4. Salesforce Churn Predictor (Difficulty: 5/10)

Build a churn classifier on telecom data, handling class imbalance and optimizing the decision threshold for business value.

What you learn: Feature engineering · Accuracy trap for imbalanced data · SMOTE · Threshold tuning & precision-recall tradeoff