rai-mldev 1.0.0

A Streamlit-based ML model development guide promoting responsible AI

ML Model Development Guide

A guided, web-based machine learning model development application built with Streamlit. This application provides a step-by-step approach to building, training, evaluating, and explaining machine learning models for both classification and regression problems, with a focus on responsible AI practices.

Overview

ML Builder is an educational and professional tool designed to guide users through the complete machine learning development lifecycle. Each stage includes built-in guidance, interactive visualizations, and ethical considerations to ensure responsible model development.

Key Features

• 9-Stage Guided Workflow: Complete ML pipeline from data loading to model explanation
• Dual Problem Support: Handles both classification and regression tasks
• Interactive Visualizations: Rich data analysis and model performance visualizations powered by Plotly
• Responsible AI Integration: Built-in bias detection, fairness analysis, and ethical considerations
• Model Comparison: Quick performance comparison across multiple algorithms
• Educational Content: Comprehensive explanations and guidance at each stage
• Export Capabilities: Download models, datasets, and complete reproduction scripts

2. Begin development by clicking "Start ML Development" and either:
   • Upload your own CSV dataset
   • Use one of the provided sample datasets (Titanic or Miami Housing)

Development Pipeline

1. Data Loading

• Secure CSV file upload with validation (100MB limit, MIME type checking)
• Sample datasets provided (Titanic classification, Miami Housing regression)
• Target variable selection with automatic problem type detection
• Basic data quality assessment

2. Data Exploration

• Multi-tab dataset analysis interface
• Statistical summaries and distributions
• Correlation analysis using Dython for mixed data types
• Missing value pattern analysis
• Interactive visualizations

3. Data Preprocessing

• Missing value imputation (mean, median, mode, advanced strategies)
• Categorical encoding (One-Hot, Label, Target encoding)
• Feature scaling (Standard, MinMax, Robust scalers)
• Data type optimization

4. Feature Selection

• Statistical feature importance analysis
• Correlation-based filtering
• Bias detection for protected attributes
• Interactive feature selection interface
• Responsible AI considerations

5. Model Selection

• Intelligent model recommendations based on data characteristics
• Quick model comparison with sample data
• Non-linearity detection using mutual information
• XGBoost compatibility checking for multiclass problems
• Educational model guide with performance characteristics

6. Model Training

• Multiple training approaches:
  • Standard training with default parameters
  • Random Search hyperparameter optimization
  • Optuna Bayesian optimization
• Cross-validation with detailed metrics
• Model enhancement (calibration, threshold optimization)

7. Model Evaluation

• Comprehensive performance metrics
• Interactive visualizations (ROC curves, confusion matrices, residual plots)
• Sample predictions display
• Model stability assessment
• Cross-validation results analysis

8. Model Explanation

• SHAP value analysis and visualizations
• Feature importance explanations
• Individual prediction explanations
• Bias and fairness assessment
• Model transparency reporting

9. Summary & Export

• Complete development journey overview
• Model recreation Python scripts
• Processed dataset downloads
• Development timeline and logging
• Environment setup instructions

Supported Algorithms

Classification Models

• Logistic Regression
• Decision Tree
• Random Forest
• Multilayer Perceptron (MLP)
• Gradient Boosting
• XGBoost (with multiclass compatibility checking)
• LightGBM

Regression Models

• Linear Regression
• Decision Tree
• Random Forest
• Multilayer Perceptron (MLP)
• Gradient Boosting
• XGBoost
• LightGBM

Sample Datasets

Two sample datasets are included:

1. Titanic Dataset (Classification)

   • Binary classification problem (survival prediction)
   • 8 features including passenger class, age, sex, fare
   • Educational dataset for learning classification concepts

2. Miami Housing Dataset (Regression)

   • Housing price prediction problem
   • 15+ features including location, size, distance metrics
   • Real-world regression problem with mixed feature types

Technical Dependencies

Core Framework

• streamlit>=1.24.0 (Web application framework)
• pandas>=1.5.0 (Data manipulation)
• numpy>=1.23.0 (Numerical computing)
• scikit-learn>=1.0.2 (Machine learning algorithms)

Visualization

• plotly>=5.13.0,<7.0.0 (Interactive visualizations)
• matplotlib>=3.6.0 (Static plotting)
• seaborn>=0.12.0 (Statistical visualization)

Advanced ML

• xgboost>=2.1.4 (Gradient boosting)
• lightgbm>=4.6.0 (Gradient boosting)
• optuna>=3.5.0 (Hyperparameter optimization)
• shap>=0.41.0 (Model explanation)

Specialized Libraries

• category_encoders>=2.6.0 (Advanced encoding)
• dython>=0.7.3 (Mixed-type correlation)
• fairlearn>=0.7.0 (Fairness assessment)
• python-magic>=0.4.27 (File type detection)
• imblearn (Imbalanced learning)
• Boruta>=0.3 (Feature selection)

Project Structure

ML_Builder_New/
├── app.py                     # Main application entry point
├── Builder.py                 # Core ML pipeline class
├── requirements.txt           # Project dependencies
├── pages/                     # Individual stage pages
│   ├── 1_Data_Loading.py
│   ├── 2_Data_Exploration.py
│   ├── 3_Data_Preprocessing.py
│   ├── 4_Feature_Selection.py
│   ├── 5_Model_Selection.py
│   ├── 6_Model_Training.py
│   ├── 7_Model_Evaluation.py
│   ├── 8_Model_Explanation.py
│   └── 9_Summary.py
├── components/                # Reusable components
├── utils/                     # Utility functions
├── content/                   # Educational content
├── sample_data/              # Sample datasets
└── docs/                     # Comprehensive documentation

Security Features

• File Upload Validation: Multi-layer security including file size limits, MIME type detection, and content validation
• Input Sanitization: Protection against code injection in column names
• Local Processing: All data processing occurs locally without external transmission
• Error Handling: Robust error boundaries prevent application crashes

Responsible AI Features

Bias Detection

• Automatic identification of potentially protected attributes
• Statistical bias testing across demographic groups
• Feature correlation analysis for fairness assessment

Model Interpretability

• SHAP value analysis for prediction explanations
• Feature importance rankings and visualizations
• Individual prediction breakdowns
• Model decision boundary analysis

Ethical Guidelines

• Stage-specific ethical considerations and prompts
• Best practice recommendations
• Impact assessment guidance
• Fairness metric calculations

Educational Components

• Interactive Model Guide: Detailed explanations of algorithm characteristics
• Performance Metrics Education: Clear explanations of evaluation metrics
• Statistical Concept Explanations: Educational content for statistical tests and measures
• Contextual Help: Stage-specific guidance and tips
• Journey Tracking: Complete audit trail of development decisions

Usage Tips

• Data Preparation: Ensure CSV files have clear headers and proper formatting
• Problem Type: The application automatically detects classification vs regression problems
• Model Selection: Use the quick comparison feature to identify promising algorithms
• Hyperparameter Tuning: Start with Random Search before using Optuna for final optimization
• Evaluation: Review all metrics and visualizations before proceeding to explanation
• Export: Use the Summary page to download complete project artifacts

Performance Considerations

• File Size Limit: 100MB maximum for uploaded CSV files
• Sample-Based Analysis: Large datasets are automatically sampled for quick model comparison
• Memory Management: Efficient data handling with automatic cleanup
• Parallel Processing: Multi-core support for compatible algorithms

Troubleshooting

Common Issues

• Upload Failures: Verify CSV format and file encoding (UTF-8 recommended)
• Model Training Errors: Check for missing values or incompatible data types
• Memory Issues: Reduce dataset size or use data sampling options
• XGBoost Compatibility: For multiclass problems, ensure class labels start from 0

Getting Help

• Review comprehensive documentation in the docs/ directory
• Check application logs and error messages
• Verify all dependencies are correctly installed
• Ensure Python version compatibility (3.8+)

Contributing

Contributions are welcome! Areas for contribution include:

• New algorithm implementations
• Enhanced visualization components
• Additional evaluation metrics
• Documentation improvements
• Bug fixes and performance optimizations

License

This project is licensed under the MIT License.

Author

Richard Wheeler
ML Engineer & Responsible AI Researcher

---

Version 1.0.0 | A comprehensive tool for responsible machine learning model development