# Predicting Problematic Internet Use in Children and Adolescents

This solution achieved a bronze medal in the Child Mind Institute - Detect Early Risk of Depression and Self-Harm competition on Kaggle. The code implements an ensemble approach to predict problematic internet usage based on physical activity data.
Key features:
Data preprocessing and feature engineering
Ensemble of machine learning models (XGBoost, LightGBM, CatBoost)
Custom metric optimization using quadratic weighted kappa
Cross-validation with stratified k-fold
Threshold optimization for improved predictions
The solution provides a comprehensive approach to the competition problem, incorporating various machine learning techniques and custom optimizations to achieve competitive results
[https://www.kaggle.com/code/razvanmihaihanghicel/predicting-problematic-internet-use-with-xgboost](https://www.kaggle.com/code/razvanmihaihanghicel/predicting-problematic-internet-use-with-xgboost)