Time Series Forecasting

Our second approach to forecasting is called Time Series Analysis, and we’re going to use this when we have time-series data.

Time series data is simply historical data that’s date and time stamped at regular intervals. We can look at trends over time and then we can infer that those trends are likely to continue.

In the above image, we have time-series data from 1946 – 1959 of the average births per month in a certain city.

We’re able to plot these births on a graph:

We can clearly see an overall trend of births increasing year-by-year, but we can also see seasonal trends within each year.

We can decompose this time-series graph in several different ways, for example, the initial data, the overall trend, the seasonal trend and random noise:

This helps us to project our trends into the future like this:

Note the purple shaded area shows some uncertainty based on the difference between the random noise and the trend line.