Time series are among the most common shapes of business data, because almost everything is recorded with a timestamp.
What makes them special is that the order of the rows carries information.
Methods that treat rows as independent miss that information and often give misleading answers.
A time series has a time index, a value, and a frequency such as hourly, daily or monthly.
Irregular events, such as individual orders, are usually totalled into regular periods first.
In pandas, a datetime index and the resample method turn timestamped events into a regular series.
Missing periods should appear as gaps, not be silently skipped, or the frequency breaks.
Duplicate timestamps, often from loading the same day twice, double the values for those periods.
Time zones and daylight saving changes can create missing or repeated hours in hourly data.
Most series combine a trend, seasonal patterns and irregular noise.
Separating these parts, called decomposition, shows each one clearly.
Events such as holidays, outages and promotions create spikes that are not part of the normal pattern.
Recording these events alongside the series explains the spikes and stops them distorting forecasts.
In practice, plotting the series is always the first step, before any model.
The common mistake is splitting or shuffling time series data as if the rows were independent.