To propose a stepwise method for Responsible Data Selection (ReDS) for algorithmic personalization of mHealth, addressing ethical and regulatory implications of user input data.
Approach:
Step 1: Identify the personalization objective, which is to personalize the type of challenge to promote adherence and diversify coping strategy types.
Step 2: Identify promising input data, such as the emotional state of adolescents and prior completion rates.
Step 3: Acknowledge the ethical implications of using sensitive personal emotion data.
Step 5: Analyze the utility of all data features using evaluative simulations with reinforcement learning models.
Step 6: Determine the utility-risk trade-off, concluding that tiredness data can be used as an alternative to emotion data with risk mitigation strategies.
Key Findings:
Using solely the completion rates of the previous day can benefit the personalization objective.
Incorporating tiredness data can further enhance the performance of the personalization algorithm.
The ReDS method provides a practical framework for integrating ethical considerations in algorithm development.
Interpretation:
The study demonstrates the practical utility of the ReDS method in addressing ethical concerns in data selection for personalization algorithms in mHealth.
Limitations:
The study focuses on a specific case study involving adolescents, which may limit generalizability.
The effectiveness of risk mitigation strategies for sensitive data was not empirically tested.
Conclusion:
The work aims to inspire future developers of personalization algorithms to incorporate ethical considerations explicitly in their development processes.
