Linear Mixed Models of ROC-AUC for Smoking Lapse Prediction

A study from 2026 analyzed data from 37 participants who completed up to 16 ecological momentary assessments per day for 10 days during a smoking cessation attempt. The dataset contains results from random forest algorithms predicting smoking lapses and cravings, evaluated via F1-scores and ROC-AUC using mixed effects models. It was authored by Corinna Leppin and shared on figshare under a CC-BY-4.0 license.

Use Cases

• Optimizing EMA prompt frequency for predictive models based on reported performance trade-offs
• Comparing feature reduction strategies for lapse and craving prediction algorithms
• Evaluating the impact of including participant-specific training data on model performance
• Benchmarking machine learning model performance (F1-score, ROC-AUC) for behavioral prediction tasks

Strengths

• Performance metrics are provided with interquartile ranges (IQR), offering a view of variability
• The study systematically tests multiple configurations: EMA frequency (3 to 16 prompts/day), predictor count, and training data source
• Results are disaggregated for two distinct prediction targets: lapses and cravings

Limitations

• The dataset is very small at 5.5 KB, suggesting it contains summary model results, not the raw participant assessment data
• Row count and column-level documentation are absent, limiting understanding of the data structure
• The description notes 'modest overall performance and substantial inter-individual variability' for the predictive models

Provenance

Source

figshare

Collection Method

Data likely originates from a research study where 37 participants completed ecological momentary assessments (EMAs) reporting mood, context, behavior, cravings, and smoking lapses.

Time Range

The study period covered the first 10 days of participants' smoking cessation attempts.

Freshness

Last updated 2026-05-14 17:33:17