Data on social phenomena can fluctuate significantly, especially in the internet age. New trends, memes and socio-political movements come and go. This is reflected in internet traffic. Websites number of visits shows impulsive changes that overlap with long-term trends and seasonal patterns.
In this article, I show you how to forecast time series that is quite unpredictable by its nature. You will learn how to use weather data to predict human activity, specifically online. I will show you how to improve your predictions using the domain knowledge of the target variable. We will briefly discuss feature engineering and the evaluation of models. Finally, we develop a web app that trains and tests several predictive models and evaluates them.