An automated machine learning pipeline that profiles datasets, preprocesses data, selects features, trains and compares multiple models, and generates a professional evaluation report, all without requiring an API key.
Try it live: eugen-goebel-predictive-analytics-agent-app-l05zcc.streamlit.app. Click "Use sample dataset (Customer Churn)" and the full ML pipeline runs in your browser: profiling → preprocessing → training → evaluation, with no API key required.
Data Analysis: auto-detected target column, task type, and preprocessing pipeline
Model Comparison: cross-validated accuracy, standard deviation, and training time across 4 models
Evaluation Results: test/train scores, overfitting detection, and accuracy comparison chart
- Auto-Detection: Automatically identifies the target column and task type (classification or regression)
- Data Profiling: Analyzes data quality, distributions, missing values, and column statistics
- Smart Preprocessing: Handles missing values, encodes categoricals, and scales features
- Feature Selection: Applies variance thresholding and statistical feature selection (SelectKBest)
- Model Comparison: Trains 4 models with 5-fold cross-validation and selects the best
- Evaluation: Generates confusion matrices, feature importance charts, model comparison plots, and overfitting detection
- Report Generation: Creates a professional DOCX report with all results and visualizations
- Web Interface: Interactive Streamlit app for uploading data and exploring results
- No API Key Required: Runs entirely locally using scikit-learn
The project uses a multi-agent architecture where each agent handles one pipeline phase:
MLPipelineOrchestrator
├── DataProfiler → Dataset analysis & target detection
├── PreprocessorAgent → Cleaning, encoding, scaling
├── FeatureEngineerAgent → Feature selection & ranking
├── ModelTrainerAgent → Training & cross-validation (4 models)
├── EvaluatorAgent → Metrics, charts, overfitting detection
└── ReportGenerator → Professional DOCX report
Classification:
- Logistic Regression
- Random Forest Classifier
- Gradient Boosting Classifier
- K-Nearest Neighbors
Regression:
- Linear Regression
- Random Forest Regressor
- Gradient Boosting Regressor
- K-Nearest Neighbors Regressor
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt# Run with sample dataset
python main.py
# Run with your own data
python main.py path/to/your_data.csv
# Specify output directory
python main.py data.csv --output reports/streamlit run app.pyUpload a CSV/Excel file or use the built-in sample dataset. The app displays:
- Data profile with quality metrics
- Preprocessing steps applied
- Model comparison table with scores
- Evaluation charts (confusion matrix, feature importance, etc.)
- Download button for the full DOCX report
Includes a customer churn dataset (data/sample_customers.csv) with 80 rows and 11 features:
| Feature | Description |
|---|---|
| age | Customer age |
| income | Annual income |
| credit_score | Credit score |
| years_customer | Years as customer |
| num_products | Number of products |
| has_mortgage | Has mortgage (0/1) |
| has_online_banking | Uses online banking (0/1) |
| monthly_charges | Monthly charges |
| total_charges | Total charges |
| support_calls | Number of support calls |
| churn | Target: whether customer churned (0/1) |
pytest tests/ -v35 tests covering all agents and the end-to-end pipeline.
predictive-analytics-agent/
├── agents/
│ ├── data_profiler.py # Dataset profiling & analysis
│ ├── preprocessor.py # Data cleaning & transformation
│ ├── feature_engineer.py # Feature selection & ranking
│ ├── model_trainer.py # Model training & comparison
│ ├── evaluator.py # Model evaluation & charts
│ └── orchestrator.py # Pipeline coordinator
├── utils/
│ └── report_generator.py # DOCX report generation
├── tests/
│ ├── test_profiler.py
│ ├── test_preprocessor.py
│ ├── test_feature_engineer.py
│ ├── test_model_trainer.py
│ ├── test_evaluator.py
│ └── test_pipeline.py
├── data/
│ └── sample_customers.csv
├── app.py # Streamlit web interface
├── main.py # CLI entry point
└── requirements.txt
- scikit-learn: Machine learning models, preprocessing, evaluation
- pandas: Data manipulation
- matplotlib: Chart generation
- Streamlit: Web interface
- python-docx: Report generation
- Pydantic: Data validation with typed models
MIT