Time series, as the name suggests, is data that contains certain time periods or time stamps. It contains observations over certain time period. This kind of data tells us about how variables change over time based on various factors. Time series analysing and forecasting can be used to predict data with respect to some future time.
Univariate time series contains values taken with respect to a single variable at certain time instances over a period of time. A multivariate time series contains values taken with respect to multiple variables at same periodic instances of time.
Time series consists of 4 components that have been described below −
It refers to the mean of the values of the data around which the values vary.
As the name suggests, it tells about a certain pattern (increasing, decreasing, neutral, and so on) of the data point in the data set with respect to the time.
It tells about the repetitive nature/cyclic behaviour of the time series data.
The unwanted values in the observations that are added implicitly due to external factors.
There are many methods with respect to time series that can be used to capture the above mentioned components. Let us understand a few of them in brief −
Simple estimation techniques can be used to predict value of a given variable when its previous time’s actual value has been provided. This is not really used on its own, but more like a reference point while working with sophisticated techniques.
Auto−regression is a technique that predicts values for future period of time. It uses historical data, i.e previous time period data. The data may fit the model better, but auto−regression doesn’t provide accountability for the ‘seasonality’ component.
ARIMA stands for ‘Auto Regressive Integrated Moving Average’. It tries to predict the value of a data point as a linear function of historical/previous data points. This way, the errors present in the previous times would get incorporated into the predicted data. In real−life situations, the data may not be stationary, due to which, Seasonal ARIMA and Fractional ARIMA have been developed.
Exponential smoothing is a technique in which the variable is modelled as an exponential weighted linear function of the previous values in the dataset. This kind of model works well with trend and seasonality of the data.
LSTM refers to Long Short−Term Memory model, which is a recurrent neural network that is used to work with time series model when there are long term dependencies. Large amounts of training data is required to gather the trends in multi−variate time series dataset.