About the project
A multiclass project that predicts the flight class (Business, Eco or Eco Plus) on the Airline Passenger Satisfaction dataset from Kaggle. The main model is a multilayer perceptron built from scratch in PyTorch, with a scikit-learn Random Forest as the reference. Preprocessing is fitted on the training split alone, hyperparameters come from an 18-configuration grid search on a dedicated validation split, and the test split is touched exactly once at the end.
What it demonstrates
- MLP from scratch in PyTorch
- 18-configuration grid search
- Random Forest reference
- No data leakage by construction
- Minority-class handling without resampling
Stack
- Python
- PyTorch
- scikit-learn
- pandas
- Streamlit
