Marwan_Portfolio

This page summarizes all the projects I worked on, if you want to explore a project more just click the title !

Project 1: Churn Prediction classifier

• Created a Machine learning model that solves a churn classification problem (IBM Data Set).
• Performed a lot of data exploratory analysis to better understand the customer segments with respect to the target variable churn.
• Engineered features (dummy variables, aggregate monthly and total charges per servies).
• Balanced the data using oversampling & undersampling techniques (SMOTE).
• Improved the base model performance from 67% recall for minority group and 81% accuracy to 97% for both accuracy and recall.
• Built a user facing API using streamlit, where user/company employee can upload a csv file of the data set and get the same dataset with a churn_probability column.

Project 2: Flight Delays Prediction Estimator

• Created a Machine learning model that estimates flight delays
• Wrote SQL queries in VSCode to export a sample of the data stored in a Postgress database
• Engineered features like weahter conditions (using weather API),route traffic, and aggregate datetime dates.
• Optimized our linear regression model performance by using ensemble stacking technique to get best model.
• In process of building a user facing API using streamlit

Project 3:Loan Automation Approval Process (Mini Project, 1 day)

• Created a Machine learning model that predicts the probability a cutomer will get approved for a loan or not based on a data set with 11 features.
• Cleaned the data properly (missing values, extreme values)
• Performed a lot of data exploratory analysis to better understand the customer segments with respect to the target variable.
• Optimized the base model accuracy from 69% in terms of accuracy and 78% for F1 score to 81.3% in terms of accuracy and 88% F1 score using piplines grid search to hyperparamter tune and cross validate.
• Deployed the model using Flask & AWS
• In process of building a user facing API using streamlit to take the entire csv file and return a csv file with a prediction probability column.