I'm Yahya

The Titanic competition is based on the infamous shipwreck of Titanic in April 15, 1912. The goal is to create a model that predicts which passengers survived the Titanic shipwreck.

• I did a detailed analysis of the input features in order to understand what impact does each feature has on the target.
• After selecting suitable features, I developed strategies to fill in missing values and created suitable encoding schemes for non-numerical features.
• I developed a complete pipeline to automatically perform the necessary data preprocessing.
• Then, I tested a lot of commonly used classification models. From which, I found out that Support Vector Machine Classifier and Random Forest Classifer are the most promising.
• Finally, I ran grid searches on these two classifiers to find the best set of parameters. The best model achieved roughly 82% accuracy on cross-validation, and I used this model to predict which passengers survived in the test set.

The goal of the bank churn competition is to predict whether a customer continues with their account or closes it (churns).

• I did a detailed analysis of the input features in order to understand what impact does each feature has on the target.
• The dataset had no missing values, however, there were duplicate rows and some non-numeric columns. I dropped the duplicate rows and one-hot encoded the non-numeric columns.
• To make the data preprocessing flexible and simple to use for new data, I created a complete pipeline to automatically perform the necessary steps.
• Based on the nature of the dataset, a decision tree ensemble model was the most suitable. For it, I used the LightGBM framework.
• The best model was evaluted based on the ROC AUC criteria. It achieved a public score of roughly 0.75 on Kaggle, and I used this model to predict whether the customer will churn or not.