Knowledge Base & Community Wiki
Time Series Forecasting – Tutorial I
What is Time Series Forecasting – Time Series Forecasting involves the use of various modelling approaches e.g. ARIMA, ETS, etc. for purposes of forecasting the behaviour of the series into the future. Modelling algorithms for Time Series Forecasting consume Time Series data as input and provide Time Series Forecasts as output. Auto-Regressive Integrated Moving Average (ARIMA) models and Exponential Smoothing (ETS) models provide two different approaches to time series forecasting. Exponential Smoothing (ETS) and ARIMA models are the two most widely-used approaches to time series forecasting, and provide complementary techniques for purposes of Time Series Forecasting. While exponential smoothing models were based on a description of trend and seasonality in the data, ARIMA models aim to describe the autocorrelations in the data.
In statistics and econometrics, an Auto-Regressive Integrated Moving Average (ARIMA) model is very commonly used technique for purposes of forecasting. These models are fitted to time series data either to better understand the data or to predict future points in the series i.e. for purposes of forecasting. ARIMA models are applied in some cases where data show evidence of non-stationarity, where an initial differencing step (corresponding to the “integrated” part of the model) can be applied to reduce the non-stationarity. The forecasting capability in VisualizeIT offers a combination of both ETS and ARIMA forecast models.
Exponential Smoothing (ETS) is a very popular approach to produce a smoothed Time Series. Whereas in Single Moving Averages the past observations are weighted equally, Exponential Smoothing assigns exponentially decreasing weights as the observation get older. In other words, recent observations are given relatively more weight in forecasting than the older observations. In the case of moving averages, the weights assigned to the observations are the same and are equal to 1/N. In Exponential Smoothing (ETS), however, there are one or more smoothing parameters to be determined (or estimated) and these choices determine the weights assigned to the observations.
Non-seasonal Auto-Regressive Integrated Moving Average (ARIMA) models are generally denoted ARIMA(p, d, q) where parameters p, d, and q are non-negative integers, p is the order of the Autoregressive model, d is the degree of differencing, and q is the order of the Moving-average model. Seasonal ARIMA models are usually denoted ARIMA(p, d, q)(P, D, Q)_m, where m refers to the number of periods in each season, and the uppercase P, D, Q refer to the autoregressive, differencing, and moving average terms for the seasonal part of the ARIMA model. ARIMA models form an important part of the Box-Jenkins approach to time-series modelling.
When two out of the three terms are zeros, the model may be referred to be based on the non-zero parameter, dropping “AR”, “I” or “MA” from the acronym describing the model. For example, ARIMA (1,0,0) is AR(1), ARIMA(0,1,0) is I(1), and ARIMA(0,0,1) is MA(1).
What is the Time Series Forecasting capability within Visualize IT – The two Time Series Forecasting algorithms implemented within VisualizeIT are ETS and ARIMA. Let’s dig a bit into both a bit deeper. Time Series Forecasting within VisualizeIT allows the use of ETS (Exponential Time Series) and ARIMA Forecasting techniques for purposes of obtaining forecasts. This section provides an overview of the A pre-requisite for using the Time Series Forecasting functionality within VisualizeIT is that you have uploaded data for the given Application / Data Dimension using the Data Management functionality within VisualizeIT. With the data uploaded you are then able to use the Time Series Forecasting functionality to explore your data.
Let’s start by taking a look at the approach to using the Time Series Forecasting functionality within VisualizeIT. To get started you would need to log-in to VisualizeIT, select the Statistical Modelling link from the Left Hand Side menu and then select Time Series Forecasting. You are then presented with the Time Series Forecasting form. Step 1 of the Time Series Forecasting form requires selection of the Application including associated Data Dimension.
Step 2 of the Time Series Forecasting form requires selection of the time period i.e. Start Time, End Time including the Roll-up you would like to perform. Please note that for large data sets i.e. Data sets spanning years you will need to perform a Roll-up of your data before sending it through for purposes of Forecasting –
- Example 1 – Application / Data Dimension with data sampled at 5 min intervals (1 record every 5 minutes) with data collected for a period of 6 months. When using the Time Series models to obtain forecasts for a period of 1 month, you will need to roll up this data to at-least, “1 Hour”. If possible roll it up to, “1 Day”.
- Example 2 – Application / Data Dimension with data sampled at Daily intervals (1 record every day) with data collected for a period of 5 Years. When using the Time Series models to obtain forecasts for a period of 1 year, you will need to roll up this data to at-least, “1 Week”. If possible roll it up to, “1 Month”.
If you choose not to Roll-up your data you will be limited by the range of “Forecast Horizon” options that are available to you. Please make sure you perform the relevant Roll-ups of your data set on Step 2 of the Time Series form. When using data sets collected at granular intervals over a large time period we would highly recommend using the Roll-up option to obtain lengthy “Forecast Horizon” options.
Step 3 of the Time Series Forecasting form requires you to select the “Periodicity” within the data set along with choice of the Time Series Forecasting technique i.e. ETS or ARIMA and most importantly the “Forecast Horizon”. As mentioned earlier, the choice of “Forecast Horizon” for your given Application / Data Dimension depends on the Roll-up options you’ve (or not) selected.
Having selected the relevant options you then hit the “Visualize Data” button on Step 3 which then sends the input parameters to the modelling engine and generates the relevant forecasts for the given Application / Data Dimension. Please do keep in mind that Time Series Forecasting is compute intensive and you should give the system 20-30 seconds after submitting the input values which is what it should usually take for the resulting forecasts to appear.
Conclusion – Time Series Forecasting capability within VisualizeIT provides you the ability to obtain Time Series Forecasts using the ETS or the ARIMA modelling techniques. Both ETS and ARIMA are well known industry standard Time Series Forecasting techniques. We suggest using caution when using the Time Series Forecasting techniques because the past is not always a great predictor of the future. We also encourage you to make use of the Time Series EDA (Exploratory Data Analysis) and Time Series Decomposition modelling functionality within VisualizeIT to understand your data set for the given Application / Data Dimension before you go consider using the Time Series Forecasting capability.
Modelling Solution: VisualizeIT offers access to a bunch of Analytical Models, Statistical Models and Simulation Models for purposes of Visualization, Modelling & Forecasting. Access to all the Analytical (Mathematical) models is free. We recommend you try out the Analytical models at VisualizeIT which are free to use and drop us a note with your suggestions, input and comments. You can access the VisualizeIT website here and the VisualizeIT modelling solution here –VisualizeIT.