cczarnuch
commented
4 years ago With some of the data we could try to use stats models like ARIMA. I've used more advanced stats models like Holt Winter’s Exponential Smoothing for forecasting as well. For more advanced methods we could consider LSTMs or other Seq2Seq type models.
We could also consider doing anomaly detection on the data (might be difficult since this would have to be done unsupervised), or use other methods to determine specific time windows that are of interest from the COVID patient data.
I have done some research on LSTMs and I think that would be a good option as we can use an LSTM for both forecasting and anomaly detection. Not the direction we have to go, but just a suggestion.
I was thinking for the time window we could have a 'weather report'-like view that would show the forecast for the coming week.
With some of the data we could try to use stats models like ARIMA. I've used more advanced stats models like Holt Winter’s Exponential Smoothing for forecasting as well. For more advanced methods we could consider LSTMs or other Seq2Seq type models.
We could also consider doing anomaly detection on the data (might be difficult since this would have to be done unsupervised), or use other methods to determine specific time windows that are of interest from the COVID patient data.