Web-App-Titanic-Survival

Come and check your chances of surviving the titanic shipwreck in this web app

Project Overview

Predicting if you will survive the titanic or not

• Created an accurate model that can predict the probability of you surviving or not the shipwreck
• Enter your passenger details in the web app and find out
• Predicts correctly with an 84% accuracy
• Feature engineered the titles from the passenger names
• Feature engineered if the person is alone or not from the number of relatives present on the ship
• Cleaned the data, normalized and scaled it appropriately
• Optimized Naive bayes, Logistic regression, decision tree, k nearest neighbors, random forest, support vector machine, xtreme gradient boosting using ensembling methods to reach the best model.
• Finally a soft voting ensembling classifier achieved the best accuracy.

Code and Resources Used

Python Version: 3.10.5
Packages: pandas, numpy, sklearn, requests, dill, Flask, xgboost, gunicorn, matplotlib, seaborn

Data from the passengers:

Variable	Definition	Key
survival	Survival	0 = No, 1 = Yes
pclass	Ticket class	1 = 1st, 2 = 2nd, 3 = 3rd
sex	Sex
Age	Age in years
sibsp	# of siblings / spouses aboard the Titanic
parch	# of parents / children aboard the Titanic
ticket	Ticket number
fare	Passenger fare
cabin	Cabin number
embarked	Port of Embarkation	C = Cherbourg, Q = Queenstown, S = Southampton

EDA

After getting the data, I explored it and looked for correlations:

• Plotted the relationship between the features and the target variable
• Compared various features to one another
• Determined whether or not features are unbalanced. Checked if the target's classes are unbalanced
• Calculated correlations between the various columns

Feature engineering and cleaning

The steps I took in this phase:

• Drop out PassengerId(irrelevant), Name(feature engineered), Ticket(irrelevant) and Cabin(irrelevant and too many NaNs).
• Create a new feature 'IsAlone'( = SibSp + Parch + 1) to indicate if a passenger is alone.
• Extract titles from Name.
• Imputation of missing values and normalization of numerical features
• Encode the categorical features.

Model Building

I just used 'Age', 'Fare', 'Pclass', 'Sex', 'Embarked', 'IsAlone', and 'Title' because according to EDA they were the most relevant.
Created a numerical pipeline, then a categorical pipeline, and then united them.

Then I passed the features through the pipeline
Then I applied several machine learning models to the data and computed their cross validation scores on a validation set

I plotted the learning curves for all of those models to see who would overfit or underfit
I finally settled on an SVM, tuned it using gridsearchcv and computed its accuracy

Model Additional Ensemble Approaches

Here I used ensembling algorithms to try to improve the model(Due to computational reasons, I did not tune these ensemble models to their max. This explains why their accuracy may be lower than that of the tuned SVM model)

1. Experimented with a hard voting classifier of three estimators (KNN, SVM, RF) (81.4%)

2. Experimented with a soft voting classifier of three estimators (KNN, SVM, RF) (81.7%) (best performance in competition leaderboard)

3. Experimented with soft voting on all estimators performing better than 80% except xgb (KNN, RF, LR, SVC) (82.6%)

4. Experimented with soft voting on all estimators including XGB (KNN, SVM, RF, LR, XGB) (82.8%) (Best Performance)

Try the app