thanks for using my API, yes there are lot differences.

1. Prophet and Neural Prophet they share the same syntax, Thus gives a boost in faster implementation with less coding experience.

2.1 XGBoost, RF, DT or any classifical ML is not the option because they can predict only t+1 and not t+2,3,4,5,6.......n days. Try it out since it will be a supervised learning you can feed t+1 then t+2 and so on to model and compare u will see a hetroscadicity difference with time and variance.

2.2 Additionally tons of unnecessary coding is required like series to supervised, differencing etc. just only to get t+1

2.3. If you wish to get t +2,3,4,5,.......n days u will need to convert it the target variable to (t - n) which you need to do it in LSTM. LSTM and Neural PRophet both are deep learning But why they developed NeuralProphet is LSTM is all in one winner? LSTM takes time to training and requires custom modifications each and evrytime in training to fit the data.

3. Real life scenario with Big Data it will not be energy efficient to run LSTM for what to get the same requests as NeuralProphet.

I hope that answers! Happy Machine Learning