Time Series Analysis
03. SARIMA Models
Summary:
In this exercise we'll train and score our first SARIMA models.
Instructions:
1) Run the workflow up through the Missing Value node. We’ll start this exercise from here
2) Partition the data using the Table Partitioner node. Let’s use an 80/20 split. Make sure you check the box to take data from the top. This is important with time series data.
3) Apply the SARIMA Learner to the top output from the Table Partitioner node. Note that that the SARIMA node can take some time to run so keep the settings low for now. AR and MA orders should be 2 or lower.
4) Use an SARIMA Predictor node after the learner. You can configure the number of values you want to forecast here. (hint: 7 days)
5) Replace RowIDs of the output of the SARIMA Predictor node with new RowIDs with a RowID node. Do the same for the test partition of the dataset (hint: the bottom output of the Table Partitioner node).
6) Combine the renumerated SARIMA Predictor output and the test data from the previous step with a Joiner node. Match according to RowID. Retain the Forecast column from the SARIMA Predictor output. Retain the time stamp (row ID column) and the original data (cluster_26 column) from the test data.
7) Compare the prediction vs original data with a Numeric Scorer node. Also, plot the prediction and the original data with a Linear Plot node.
8) Analyze the residuals of the SARIMA model with the Residual Analyzer node. What can you say about the residuals?
