Prophet for predict hourly basis daily temperature

[swagat123]

Hi

I have 3 months data of temperature got from censor on hourly basis. So daily 24 data points and weekly 168 data points. I need to predict next day(24 data points).

I used the below code

model = Prophet()
model.fit(df)
future = model.make_future_dataframe(periods=24, freq='H')
fcst = model.predict(future)

then I split the data into train and test and did the prediction .I got the below details Prediction quality: 101.46 MSE (10.07 RMSE) MAE - 8.30542470861 MAPE - 10.7460364749

I need to improve this forecast result.Is there any way to do this?

[vhpietil]

Most probably there are ways to improve your forecast results. But it is extremely difficult to give any suggestions without seeing your data and predictions. Someone might be able to help if you attached few plots

[swagat123]

data_mnthly.txt

I have added the screen shot along with data. I added prediction period 72 for 3days. I split the data into train and test then check for validation.

[vhpietil]

I can't see any easy way to improve your forecasts. The data seems to be pretty much random without strong daily seasonality. I don't know how to improve the predictions if there are no external regressors that explain the intraday variability.

[swagat123]

@vhpietil Thank you for the analysis.So there is no way i can improve this model. I have one question if the voltage contributing to temperature variability , is there any way we can add that voltage value as external regressor in prophet and predict next 2/3 days temperature. if so then how we can do this.

[vhpietil]

You can use add_regressor -function to add voltage value as an external regressor.

https://facebook.github.io/prophet/docs/seasonality,_holiday_effects,_and_regressors.html

[ghost]

@swagat123: quick question: the ACF en PACF plots are not part of the fbprophet package, or are they?

[swagat123]

@WouterTromp: I created the ACF and PACF using statsmodels.graphics.tsaplots.

[ghost]

Ah great. Those I know :)