UNICEF Machine Learning Conflict Prediction
Published:
Tech Stack: Python, Pandas, Plotly, NumPy, SciPy, Scikit-Learn
A group project where I worked with data from UNICEF and Dr. Schwartz using statistical methods like hypothesis testing, correlation estimation, linear regression, classification to predict conflict worldwide and assess the machine learning models, (XGBoost, FFNN and Transformer).
Implemented a forward selection step-wise regression model which we used to determine which UNICEF/FSI-based data subsets negatively affected the accuracy of the models.
Developed easy-to-understand visualizations for a non-technical audience and tided data set to minimize errors in analysis.
Created a new model for UNICEF with 88% accuracy that minimizes overfitting, accurately predicting real-world conflict.
Assessed ethical implications relevant to the current efforts of UNICEF to minimize global conflict and also protect the global security and privacy of people from data collection threats.