Table 2.
Prediction RMSE for proposed model and baselines on different days of the next cycle for the full menstruator dataset (averaged over 186 106 users)
| Model | Day 0 | Day 14 | Day 21 | Day 28 | Day 30 | Day 40 |
|---|---|---|---|---|---|---|
| Mean | 7.50 | 7.43 | 7.29 | 7.81 | 8.99 | 21.92 |
| Median | 7.49 | 7.43 | 7.32 | 7.99 | 9.35 | 23.39 |
| CNN | 8.03 | 7.97 | 7.85 | 8.23 | 9.55 | 24.51 |
| LSTM | 7.40 | 7.34 | 7.20a | 7.72a | 8.98 | 22.68 |
| RNN | 7.76 | 7.70 | 7.56 | 7.92 | 9.07 | 22.95 |
| Proposed model (predict with s = 0) | 7.56 | 7.51 | 7.36 | 7.80 | 8.59 | 14.78 |
| Proposed model | 7.38a | 7.32a | 7.22 | 7.93 | 8.58a | 11.77a |
Our model typically outperforms summary statistic-based and neural network–based baselines when we account for skipped cycles.
CNN: convolutional neural network; LSTM: long short-term memory; RMSE: root mean squared error; RNN: recurrent neural network.
a
Best-performing model on each day.