Repository contains simple model for predicting survival of passengers on Titanic Spaceship competition on kaggle [https://www.kaggle.com/competitions/spaceship-titanic]
For now there are two approaches being used:
- Deep learning model based on classical linear layers
- Machine learning models which include:
- Random Forest Classifier
- LGBM
- KNN
Also several approaches to data preprocessing are being used:
- Dropping all nan values
- Filling nan values with mean/median
- Statistical approach combined with KNN imputations.
| Model | Best score |
|---|---|
| Linear model (DL) | 79.4 % |
| Random Forest Classifier | 79.6 % |
| LGBM | 78.6 % |
| KNN | 76.6 % |
Model is based on simple linear layers with dropout and batch normalization. It is trained using Adam optimizer and binary cross entropy loss function. Currently it consists of 3 layers with 128, 32 and 1 neuron, respectively.
[ ] Add predictor for ML algorithms [ ] Add more preprocessing methods [ ] Add more models