In the previous article, we got introduced to the concept of the Time series and the overall required preconditions. Now it is time to understand the in-depth details of Time series data. To begin with, we can explore the various components of the Time series data.
Decomposition of Time series:
Let’s take a sample dataset to understand the concepts better. We can consider any sales dataset but for our experiment, we are taking the popular Airline passenger dataset.
Any time-series data can be represented as a combination of these three blocks(Trend, seasonality, noise(irregularity)). Either it could be additive or multiplicative.
additive = Trend + seasonality + noise(irregularity)
multiplicative = Trend * seasonality * noise
Trend: The overall increase or decrease pattern in the given TS data. It can also be viewed as a very high-level change.
Seasonality: The patterns that get repeated for a certain interval. If we are accounting for yearly data, there will be some seasonality that exists every year(subdivision as months).
Noise: A random details present in the TS details which are given the least importance. It does not follow any regular patterns, that's why it is also called irregularity.
From the above explanations, we can easily conclude Trend and Seasonality can be forecasted(leaving out the noise). The objective of the time series model building is to predict the trend that is present in the input TS along with the Seasonality cycles.