Basic Concepts in Time Series Analysis (Part-5)
This post is the fifth part of a series of tutorials regarding time series analysis.
In this fifth post of the series, we will present briefly the necessary steps in order to produce time series forecasts:
- Problem identification. Sometimes this is the most difficult task for producing forecasts. In this first step it should be made very clear what would be the variable of interest for which forecasts will be produced and what methods will be used for this task.
- Data collection. In this step, a significant amount of attention should be given for the collection and maintenance of the data. It is vital to be able to exploit domain knowledge and critical thinking of the employers that will assist in understanding better the data at hand.
- Time series preparation. Aim of this step is to acquire as much information as possible from the time series representation of the data. Possible seasonality or trend must be detected as well as outliers must be treated accordingly.
- Model selection. After some critical thinking on the type of the problem as well as the available data, the next step would be to decide which forecasting methods will be used to solve the problem given the particularities of each model.
- Forecast evaluation. After some forecasts have been produced from the models we chose, it is time to measure their performance using appropriate accuracy metrics. In this step it is vital to observe how the error fluctuates by applying different models or parameter values to the problem in order to eliminate bias.