Types of Time Series

Not all time series data behaves the same way. Before analyzing or forecasting, we must understand what type of time series we are dealing with. This decision affects everything — visualization, modeling, and evaluation.

In this lesson, we will classify time series data in a clear, practical way using real-world examples and simple code.

1. Classification Based on Time Index

The most basic way to classify a time series is by how frequently data is recorded.

In practice, almost all business and ML problems use discrete time series. Continuous series are usually converted into discrete form.

2. Classification Based on Number of Variables

Another important distinction is how many values are recorded at each time step.

Most beginner problems start with univariate series. Real-world forecasting systems almost always become multivariate.

Example: Univariate Time Series

Let’s create a simple univariate time series representing daily website visits.

import pandas as pd

dates = pd.date_range(start="2023-01-01", periods=7, freq="D")
visits = [120, 135, 150, 160, 155, 170, 180]

ts_uni = pd.Series(visits, index=dates)
print(ts_uni)

2023-01-01    120
2023-01-02    135
2023-01-03    150
2023-01-04    160
2023-01-05    155
2023-01-06    170
2023-01-07    180
Freq: D, dtype: int64

Here, each date has only one value. This is the simplest form of time series.

Example: Multivariate Time Series

Now let’s extend the same example by adding another variable.

import pandas as pd

dates = pd.date_range(start="2023-01-01", periods=7, freq="D")

data = {
    "visits": [120, 135, 150, 160, 155, 170, 180],
    "ad_spend": [20, 25, 30, 35, 33, 40, 45]
}

ts_multi = pd.DataFrame(data, index=dates)
print(ts_multi)

            visits  ad_spend
2023-01-01     120        20
2023-01-02     135        25
2023-01-03     150        30
2023-01-04     160        35
2023-01-05     155        33
2023-01-06     170        40
2023-01-07     180        45

Now each time step has two related variables. This allows us to model relationships between inputs and outputs.

3. Classification Based on Behavior

Time series can also be classified based on how they behave over time.

Most real-world time series are non-stationary. That’s why later lessons focus heavily on stationarity.

4. Classification Based on Predictability

Some time series are easier to predict than others.

• Deterministic: clear pattern (e.g., sine wave)
• Stochastic: randomness dominates (e.g., stock prices)

Forecasting is about finding signal inside noise.

Where This Matters in Real Projects

Choosing the wrong model usually starts with misunderstanding the type of time series.

Practice Questions

Quick Quiz

Lesson Takeaways

• Time series can be classified in multiple ways
• Univariate vs multivariate matters a lot
• Most real data is discrete and non-stationary
• Understanding the type comes before modeling