SMARTPHONE ADDICTION PREDICTION

Kaggle Playground Series S6E8 | Personal project | 2026

Built an end-to-end binary classification pipeline for the Kaggle Playground Series competition "Predicting Smartphone Addiction". The goal is to predict whether a person is addicted to their smartphone from behavioural and demographic features, working with roughly 690,000 labelled training rows and 296,000 unlabelled test rows, where every feature contains missing values.

Key contributions:

Performed exploratory data analysis on the training set, identifying a moderately imbalanced target (about 71% addicted), missingness in every column and a small fixed category set for each categorical feature.

Designed leak-free preprocessing with scikit-learn pipelines and a ColumnTransformer: median imputation and scaling for numeric features, one-hot encoding for gender, ordinal encoding for stress level and binary encoding for academic or work impact, fitted only on the training split.

Compared Logistic Regression, Random Forest and XGBoost on a stratified 80/20 validation split using ROC-AUC, with XGBoost reaching 0.96 as the leading candidate for submission.

Tuned XGBoost with RandomizedSearchCV (3-fold stratified cross-validation) and evaluated whether the tuned configuration beat the default before choosing the final model.

Collaborators: Gabriela Słomiany

Software Used: Jupyter Notebook, scikit-learn, XGBoost, pandas, Kaggle

Programming Languages: Python

Repository: github.com/imgabrielas/Smartphone-Addiction-Analysis

Project Date: August 2026