Description: The main purpose of this project is to predict the Passenger survived or not from Titanic disaster.
- Step-1: Download dataset from https://www.kaggle.com/c/titanic/data, store it in dataframe.
- Step-2: Check for the columns which contain null values and apply feature engineering techniques as stated below:
- Calculate mean of age separately for male and female and substitue it in places were null value occurs.
- Calculate mean of fares and substitue it in places were null value occurs.
- Replace null values of Embarked as S as that occurs the most.
- Step-3: We will drop Passenger ID, Name, Cabin and Ticket from Train and Test Dataset.
- Step-4: Next we need to convert the columns Sex and Embarked into numbers. We will do it with One Hot Encoding except reduce one categorical level to reduce number of columns.
- Step-5: Lets do some more feature Engineering:
- Remove Outliers.
- Create age groups fare groups and assign them numbers and create new columns of the same and drop fare and age.
- Step-6: Now that we our done with feature engineering and data is ready to be fed to the model lets try some ML algorithm and see which is the best
- Step-7: Create a dataframe with PassengerID and Survived value predicted for the test dataset using the ML algorithm which gives the best accuracy score to submit predicted value into required submission format
- The Algorithm gives the best accuracy is Decision Tree with 90.32% for the given dataset.
| Algorithm | Accuracy |
|---|---|
| Decision Tree | 90.32% |
| Random Forest | 89.75% |
| KNN | 85.59% |
| Support Vector Machine | 83.22% |
| Logistic Regression | 80.41% |
| Naive Bayes | 78.04% |