查看文章

Objectives. We developed a survival scorecard from the Titanic passenger dataset using statistical scoring methods. We discuss the scorecard development process, assess the effectiveness of statistical scorecards, and analyze the characteristics of Titanic passengers that led to survival.

Methods. From the Titanic dataset of 1,309 passengers and a binary dependent variable representing survival, we assessed nine (9) features using chi-square, Weight of Evidence (WoE), and Information Value. A logistic regression was fitted on the feature WoEs to predict survival, feature coefficients were used to determine score weights for each attribute, and an additive model on attribute scores per passenger determined survival scores. The resulting scorecards were assessed for risk ranking, and the characteristics of the passenger population were assessed for survivability and population shifts.

Results. The resulting survival scorecard was able to rank survivability amongst Titanic passengers (KS 0.558, ROC 0.892, CAP 0.765) with survival classification accuracy varying by score cutoff (AR 0.63–0.84). Sex was the strongest predictor of survivability (IV 145), followed by fare amount (IV 53), cabin class (IV 52), and passenger class (IV 51). Women passengers had four times higher survivability compared to men (72% vs. 16%). Passengers who paid $100 or more for their trip had nearly ten times higher survivability compared to free passengers (75% vs. 7.6%). Cabin passengers had higher survivability compared to non-cabin passengers with cabin B having nearly three times higher survivability compared to non-cabin passengers (75% vs. 27%). Class 1 …