Using ANNs with Keras for the Titanic Survival Prediction Dataset
This dataset was given in a Kaggle competition:
https://www.kaggle.com/c/titanic
where, given some passenger data the algorithm will predict who will die and who will survive by using some ML algorithms.
In this case, a dense (fully-connected) ANN with two hidden frames was used. Survival is treated as a binary variable: 0 if dead and 1 if survived.
Input:
Train.csv
PassengerId - a number giving the unique ID of the passenger
Survived - survival (0 = No, 1 = Yes)
Pclass - passenger class (1 to 3)
Name - passenger full name
Sex - male or female
Age - age of the passenger
SibSp - number of siblings or spouses on board
Parch - number of parents or children on board
Ticket - ticket number
Fare - ticket fare
Cabin - (if applicable) cabin in the ship
Embarked - port of embarkation (C = Cherbourg; Q = Queenstown; S = Southampton)
Test.csv
Same as Train.csv, but no Survived column
Output:
result.csv
PassengerId - a number giving the unique ID of the passenger. Should be the same as Test.csv
Survived - survival (0 = No, 1 = Yes)
Features used were PClass (since first class was prioritized when boarding), sex (females were embarked first) and Parch (kids were also prioritized).
This got 77.9% accuracy in the kaggle Test set.