CODEWITHSUNDEEP
python-3.xmachine-learningforecastingfacebook-prophet
sgl
sgl

Reputation: 95

Which forecasting to use for high frequency multi seasonable data?

I need to make a forecast model for 10 minute call centre data. My data has a week day trend and a day time trend, so there is another trend on a Monday then Sunday for example and also the specific time interval has another trend.

Also holidays and special events will show a different trend.

Is there a good fitting model for this problem?

I am using Python

Upvotes: 1

Views: 129

Answers (1)

Lukasz Tracewski
Lukasz Tracewski

Reputation: 11397

Start with a simple model that you can understand easily, say linear model. It will be your baseline. Later you might try e.g. ARIMA, (double) exponential or bayesian structural time series (BSTS) if you feel adventurous. Don't be tempted to jump to advanced models. Simple ones are great and you'll learn to appreciate them.

Invest more time in building your features, particularly around holidays. Your model will have hard times in learning the holiday schedule for the special days that are not on fixed date (and in turn have well-defined seasonality).

Prophet is easy to use (as opposed to e.g. BSTS) and nicely documented. Still, good features are more important than any individual choice of the algorithm.

Upvotes: 1

Related Questions