This study explores the use of various Explainable AI (XAI) techniques to gain a comprehensive understanding of the impact of cannabis use on the behavioral patterns and physiological states of young adult users. The researchers leveraged popular XAI methods such as SHAP, SkopeRules, decision trees, and counterfactuals to analyze sensor data collected from participants' smartphones and wearable devices.
The SHAP analysis revealed significant changes in participants' behaviors when using cannabis, including increased activity, restlessness, and connection to Wi-Fi in noisy environments. The sleep data analysis showed that some participants experienced longer sleep durations prior to cannabis use, while others exhibited nontraditional sleep patterns.
The SkopeRules extracted interpretable decision rules that linked specific sensor data patterns, such as accelerometer readings and sleep duration, to cannabis intoxication behaviors, providing actionable insights for clinicians. The decision tree models visualized the complex relationships between factors like sleep duration, day of the week, and likelihood of cannabis intoxication.
Counterfactual explanations further highlighted the key characteristics, such as sleep quality, cell phone use, and activity levels, that had a significant impact on predicting cannabis intoxication, offering personalized insights to guide intervention strategies.
By combining these diverse XAI techniques, the researchers were able to provide a multidimensional understanding of the effects of cannabis use on individual behavior and physiology. This approach offers clinicians a transparent and interpretable way to gain insights into their patients' cannabis use patterns, enabling them to develop more targeted and effective intervention strategies.
In eine andere Sprache
aus dem Quellinhalt
arxiv.org
Wichtige Erkenntnisse aus
by Tongze Zhang... um arxiv.org 04-24-2024
https://arxiv.org/pdf/2404.14563.pdfTiefere Fragen