Titanic Survival Predictor
Goal
To predict whether a person survives the Titanic.
Process
Employed a Random Forest Classifier to predict whether a given person would survive the Titanic. The CV accuracy came to be around 83%. I worked with missing values and feature engineered 4ish features along with OOB score. Baseline performance of Forest was better among other models. Tree models are able to better handle categorical data so went ahead with Random Forest. Modeling process: 1. Imports and installs 2. Seeding 3. Data exploration: features, cat vs. cant., statistics (spearman relation, p value) 4. Missing Data: (>25%, Soft label Classifier, mean, group by, median) 5. Data Imbalance: metric changes, Oversampling, undersampling via smote 6. Data processing: Label encoding, one hot encoding, scaling, split 7. Modelling: Model, CV (Kfold, Stratified Kfold, OOF) , hyperparameter tuning (Random search CV, GridSearchCV, Optuna) 8. Evaluation: Accuracy, metrics, Feature Importances