FBprophet forecasting with anomalies detection
Published:
- FBprophet gained so much popularity in modeling seasonal time series data , it has many robust tools to handle :
- Time series withs known gaps .
- Complex Periodic phenomenons such as conditional seasons.
- Robust methods to handle outliers changes in trends changes in seasonal components , holidays and special events.
- It’s a framework based on bayesian statistics , and Fourier decompositions concepts , using smoothing techniques to estimates important patterns in the data .
- This Jupyter notebook represent a gentle presentation of the general concepts of FBprophet .
- We used the Nyc passengers dataset which can be found on this link : “https://github.com/numenta/NAB/tree/master/data/realKnownCause” , which is affected by some holiday and special events, which we will handle using Fbprophet methods to handle these special cases .
- We used Isolation forest algorithme to catch anomalies in the data and show the strength of their effect .
- The fine tuning was done by Optuna which is a bayesian optimization algorithme , to add some regularization effect . and get the right hyperparameters for our holidays effects .