Kaggle - Forecasting the optimal number of agents for a Contact Center: Facebook Prophet, InfluxDB Holt-Winter
This project was built and is maintained by Tiago Oliveira - ti.olive@gmail.com.
This work was made based on the Kaggle Days Meetup Porto 2019 - Forecasting the optimal number of agents for a Contact Center
Even I couldn't finish the Kaglle challenge on time, I took the opportunity to continue the explorations over some Forecasting (multivariate & univariate) libraries and methods. So I could learn a little bit more about time-series data strcuture and Forecasting techiniques as well.
I used Facebook Prophet for hour forecasting and I loaded the dataset into an InfluxDB instance to extract some meaning with Holt-Winter method.
Kaggle late submission Score: 29.65850 with Facebook Prophet. The winner has achieved the score 8.69781.
I intend to continue my explorations testing different libraries and methods: ARIMA in Python, Spark/PySpark + MLlib, Tensorflow/Keras.