Predict Survival on the Titanic

Quick view

Goal: it is trying to use passenger list (e.g. Name, Sex, Class, Age…etc) to classify one would servive or not.

Metrics: Accuracy

Results:

• Baseline:
  • NaiveBayes (val_acc = 0.78)
  • Logistic (val_acc = 0.79)
• Feature Engineering:
  • Have age from continuous number into categorical bins
  • Clean text feature “Title”
  • Create new feature “family size” by sum up parents and childern
  • One-hot-label to categorical features, and Standarscaler() for numericals.
• ML Models:
  • Logistic regression (val_acc: 0.829)
  • KNN (val_acc: 0.829)
  • Decision Tree (val_acc: 0.836)
  • Random Forest (val_acc:0.831)
  • SVM (val_acc:0.829)
  • XGBoost (Val_acc:0.841)

  All models has been fine-tuned by RandomSearchCV

XGBoost training and validation plot

Links

Github