*Equal taxpayers
Data from wearable sensors (e.g., heart rate, step count) can be used to model mood patterns. We characterize feature representations and modeling strategies with multimodal discrete time series data for classification of mood patterns with a large naturalistically lacking data set (n = 116,819 participants) using 12 portable data streams, with an approach in capturing periodic trends in data. Considering both performance and robustness, periodicity-based aggregate feature representations with gradient boosting models outperformed other representations and architectures studied. Using periodic features improved model performance compared to temporal statistics, and gradient boosting models were more robust to losses and changes in loss distributions than a deep learning time series model.
