Toward AI-Guided Smoking Cessation: Individualized Nicotine Addiction Modeling Using Gaussian Processes

Abstract

Cigarette smoking remains the leading cause of preventable disease and death in the United States, accounting for nearly half a million deaths annually. Given the recent rise of artificial intelligence in healthcare applications, computational assessment of smoking behaviors is a promising direction. In this study, we aim to recognize and classify addiction patterns in individual smokers' daily usage based on time series data. To this end, we leverage Gaussian process modeling to iteratively learn a function that defines a smoker's behavior as usage data is accumulated. Namely, we aim to learn weekly periodic trends in usage, and then utilize the model to predict future trends. We demonstrate that the outputted predictions resemble the actual data well and that these informed forecasts significantly outperform those of a naive prediction model with respect to accuracy. Finally, we propose strategies for utilizing these predictions for goal-setting as part of a computer-supervised gradual cessation program.