Quick Guide to implementing Prophet for Time Series Forecasting

What do you say when you get a chance to implement a Machine Learning model in a real world project. I call that being lucky :)

I recently got a chance to implement a use case where we had to predict anomalies in daily data based on past trends and also predict the degree of anomaly.

The problem was divided into two parts:

1. To forecast data points for the day based on past (365 days) data. 
2. Compare actual data point with forecast and detect anomalies

The first part involved implementing a time series forecasting model. After some exploration we zeroed in on Prophet for forecasting future data and defining the upper limit. Any datapoint that breached the upper limit was classified as an anomaly.

After some trial and error, below approach is what helped me in implementing and tuning prophet in a quick and iterative manner.

This was my first practical ML use case and I am thankful to my team members who guided me and helped me successfully implement it. 

Also this is my first attempt at creating an illustration (using Adobe Express) to convey my thoughts in an easier manner (rather than lines of boring text).

Hope this is post was meaningful and of use. If yes, do give a like. If you feel additional details are needed, please let me know in comments.