ItemConference Paper

Screen Time to Severity: Machine Learning Models for Teen Smartphone Dependency Prediction

Title
Screen Time to Severity: Machine Learning Models for Teen Smartphone Dependency Prediction
Creator
Koushik, S.; Basha, Md Shaik Amzad; Sucharitha, M Martha; Ayesha, Samreen
Description
This study presents a systematic comparison of fourteen supervised classifiers trained to predict binned smartphone addiction levels (Low/Medium/High) in a cohort of 300 teenagers, using demographic, usage, academic, and health related features. After cleaning and binning the continuous Addiction_Level score into three categories, we encoded all categorical variables and standardized inputs, then stratified into 80 % training and 20 % test splits. Our expanded model suite comprised: Logistic Regression, Gaussian Naive Bayes, K-Nearest Neighbors, Decision Tree, Random Forest, Extra Trees, AdaBoost, Gradient Boosting, XGBoost, LightGBM, CatBoost, Support Vector Machine, and a multilayer perceptron (MLP). Each classifier was evaluated on accuracy, precision, recall, macro-averaged F1-score, and multiclass ROC AUC; confusion-matrix entries were flattened into nine 'Actual_i to Pred_j' columns per model for granular error analysis. Logistic Regression achieved the highest test accuracy (98.83%) , outstanding ROC AUC (0.9982) and perfect precision in discriminating the majority class ('High' addiction), despite modest recall for minority classes. MLP followed (96.33 % accuracy, 0.9878 AUC), indicating that a shallow neural network can capture nonlinear patterns but struggles on underrepresented labels. Gradient Boosting, CatBoost, and LightGBM all exceeded 95% accuracy with strong F1-scores (?0.72-0.73) and AUCs above 0.96, demonstrating the power of tree-based ensembles on mixed data types. Simpler methods (e.g., GaussianNB, KNN, Decision Tree) performed moderately (86-91% accuracy, AUC 0.84-0.98), while AdaBoost lagged (77.5 % accuracy, AUC 0.867), suggesting sensitivity to noisy features. Confusion-matrix summaries revealed that most models rarely misclassify Low-addiction teens, but confusion arises between Medium and High classes important for targeted interventions. 2025 IEEE.
Source
2025 IEEE 4th International Conference for Advancement in Technology, ICONAT 2025;
Date
01-01-2025
Publisher
Institute of Electrical and Electronics Engineers Inc.
Subject
Adolescents; Digital Well-being; Ensemble Methods; Machine Learning; Multiclass Classification; Predictive Modeling; Smartphone Addiction
Coverage
Koushik S., Christ (Deemed to be University), Computer Science and Engineering (AI/ML), Bengaluru, India; Basha M.S.A., GITAM School of Business, Gandhi Institute of Technology and Management (Deemed to be University), Hyderabad, India; Sucharitha M.M., Christ (Deemed to be University), Department of Professional Studies, Bengaluru, India; Ayesha S., Christ (Deemed to be University), Department of Professional Studies, Bengaluru, India
Rights
Restricted Access; Hardcopy may be available in the library
Relation
ISBN: 979-833159573-9;
Format
online
Language
English
Type
Conference paper

Citation

Koushik, S.; Basha, Md Shaik Amzad; Sucharitha, M Martha; Ayesha, Samreen, “Screen Time to Severity: Machine Learning Models for Teen Smartphone Dependency Prediction,” CHRIST (Deemed To Be University) Institutional Repository, accessed June 18, 2026, https://archives.christuniversity.in/items/show/26074.

Output Formats