vettorefburana/uber-orbit-forecast

Time series forecasts using Uber's Orbit model and Facebook's Prophet

Uber-Orbit-Forecast

Time series forecasting using Uber's Orbit and Facebook's Prophet

Objectives:

• Forecast Waze daily mileage data for Italy
• Forecast average daily temperatures for Italy
• Forecast mileage data using daily temperatures as explanatory variable

Models:

• Orbit Local Global Trend (LGT), Damped Local Trend (DLT) and Exponential Smoothing (ETS)
• Facebook's Prophet
• Arima

Results:

• Prophet yields a smaller prediction error than other forecasting models
• For Orbit models, prediction error increases if temperature is used as a regressor for predicting mileage
• For Prophet, including temperature as a regressor yields a smaller prediction error

References:

• https://towardsdatascience.com/time-series-modeling-with-orbit-a38e03a2a4ea
• https://arxiv.org/pdf/2004.08492.pdf

Data sources:

• Waze mileage data: https://github.com/ActiveConclusion/COVID19_mobility
• Temperature data: https://github.com/GoogleCloudPlatform/covid-19-open-data