Related Experiment Video
Updated: May 16, 2026
14:21
Creating Dynamic Images of Short-lived Dopamine Fluctuations with lp-ntPET: Dopamine Movies of Cigarette Smoking
Published on: August 6, 2013
Optimising supervised machine learning algorithms predicting cigarette cravings and lapses for a smoking cessation
Corinna Leppin1, Jamie Brown1, Claire Garnett1,2
1Department of Behavioural Science and Health, University College London, London, United Kingdom.
Plos One
|May 14, 2026
Summary
Optimizing machine learning for smoking cessation interventions requires balancing participant burden and algorithm performance. Findings suggest simpler models with less frequent data collection may be sufficient, but often fall short of ideal performance thresholds.
Area of Science:
- Digital Health
- Behavioral Science
- Machine Learning
Background:
- Just-in-time adaptive interventions (JITAIs) leverage real-time data to support behavior change.
- Ecological momentary assessment (EMA) collects frequent self-reports, but can increase participant burden.
- Optimizing EMA frequency and algorithm parameters is crucial for effective JITAI development.
Purpose of the Study:
- To systematically evaluate the impact of varying ecological momentary assessment (EMA) prompt frequency, predictor count, and training data on machine learning algorithm performance for predicting smoking cessation risks.
- To determine the optimal balance between participant burden (via EMA frequency) and predictive accuracy for a smoking cessation JITAI.
Main Methods:
- Random forest algorithms were trained and tested using data from 37 participants undergoing smoking cessation.
- Predictive performance for smoking lapses and cravings was assessed using F1-score and ROC-AUC metrics.
- Analyses systematically varied EMA frequency (16, 4, or 3 prompts/day), predictor set size, and training data source (all vs. participant-specific).
Main Results:
- Machine learning models demonstrated modest average performance in predicting smoking lapses and cravings, with significant inter-individual variability.
- Lapse prediction generally outperformed craving prediction.
- Reducing EMA frequency improved lapse prediction F1-scores but slightly decreased ROC-AUC; craving prediction metrics declined with reduced EMA frequency.
- Feature reduction and participant-specific training data did not consistently improve performance and sometimes impaired it.
Conclusions:
- While EMA-based machine learning can detect lapse risk in smoking cessation, current performance is modest and highly variable.
- More parsimonious approaches (lower EMA frequency, fewer predictors) did not consistently underperform complex ones.
- Machine learning predictions alone may be insufficient for real-world JITAIs, suggesting integration with rule-based systems is necessary.
Related Concept Videos
01:17Drug Dependence
Medications are typically administered to achieve therapeutic effects. Some drugs can modify an individual's mood and perception, frequently resulting in various enjoyable experiences. However, this can result in drug dependency, a condition marked by continuous drug use despite potential negative consequences. Drug dependency primarily falls into two categories: psychological and physical dependence. Psychological dependence occurs when the pleasurable feelings induced by the drug...
01:03Prediction Intervals
The interval estimate of any variable is known as the prediction interval. It helps decide if a point estimate is dependable.
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...