This seminar introduces an explainable digital phenotyping approach that analyzes smartphone-based Activities of Daily Living (S-ADL) by mining in-app interaction sequences collected through routine smartphone use. Unlike previous mental health prediction models that primarily rely on statistical indicators such as overall smartphone usage or time spent on app categories, this study focuses on fine-grained behavioral patterns within apps to extract features closely linked to mental health conditions. The proposed method was validated through experiments on stress and alcohol-related cognitive decline, and it improves model robustness and reliability by minimizing feature drift and covariate shift through consistent S-ADL features used both during training and after deployment. This approach enables real-time, non-invasive mental health monitoring and allows for tailored feedback based on smartphone behavior, offering strong potential for clinical applications. This seminar is designed for medical professionals interested in moving beyond traditional self-report assessments to explore the practical integration of digital health technologies in mental health care.