Towards a Smart Smoking Cessation App: A 1D-CNN Model Predicting Smoking Events

Area of Science:

• Digital Health
• Artificial Intelligence in Medicine
• Behavioral Science

Background:

• Nicotine addiction is a significant public health issue with high relapse rates.
• Smoking behavior often becomes habitual, triggered by internal and external cues.
• Current smoking cessation apps require user-initiated input, limiting their effectiveness.

Purpose of the Study:

• To develop an intelligent system for predicting smoking events.
• To overcome the limitations of existing smoking cessation applications.
• To personalize smoking cessation support by learning individual smoker routines.

Main Methods:

• Combined Control Theory and deep learning (1D-CNN) model.
• Utilized mobile device data (motion, geolocation) to learn smoker routines.
• Compared 1D-CNN against SVM and Decision Tree models.

Main Results:

• 1D-CNN model achieved an average accuracy of 86.6% in predicting smoking events.
• Accurate prediction of smoking events was observed when users actively engaged with the app.
• Mean Squared Error for nicotine level forecasting was 0.04 (weekdays) and 0.03 (weekends).

Conclusions:

• The developed model effectively predicts individual smoking events.
• This approach offers a more adaptive and personalized smoking cessation support system.
• Integration with mobile sensing shows promise for real-time intervention in smoking cessation.