This project aims to improve the prediction accuracy of wellbeing (stress, mood, and health levels) using temporal machine learning models. We extend our previous approach using Long Short-Term Memory models and time series data from the SNAPSHOT study. In addition, we consider adaptive methods to fill in missing data with time series information. We also develop the model using modifiable behavioral features such as bedtime, and examine how these contribute to wellbeing, so that people can get better control over how to improve their personal well-being.