MIS480: Predictive Analytics ‐ 1 ‐ LAB EXERCISE #1 TIME SERIES ANALYSIS FOR GLOBAL TEMPERATURE OBJECTIVE The objective of this exercise is to use time series analysis to see the global temperature trend. ACTIVITIES  Import and prepare data  Apply data mining algorithms  Configure forecasting models  Create data visualizations  Analyze and interpret output from models  Publish results SOFTWARE PREREQUISITES  SAP Predictive Analytics 2.4 or newer version  Microsoft Excel DATA REQUIRED  Data file available on http://www.ncdc.noaa.gov/cag/time-series/global. Instructions for downloading are described later in this exercise. MIS480: Predictive Analytics ‐ 2 ‐ SCENARIO Global temperatures (along with global climate) has fluctuated over the course of earth’s history. Not just Climatologists are interested in discovering seasonality, trend and cycles in past data in order to build models for forecasting future trends. As a MIS student who are taking Predictive Analytics, we would also like to spot some of these trends using time series analysis. TIME SERIES ANALYSIS Time series analysis is a technique that analysts use to (a) uncover any implicit structure (patterns or trends) in the data and (b) model that structure to make forecasts. The assumption is that the future, at least in the short term, will continue the structure of the past. This technique is useful wherever forecasting values such as sales quantities, airline passenger volume, economic metrics, and traffic volume is needed. Global temperatures (along with global climate) has fluctuated over the course of earth’s history. Climatologists are interested in discovering seasonality, trend and cycles in past data in order to build models for forecasting future trends. The interplay between earth’s atmosphere, oceans and lands poses a complex fluid mechanics problem which is governed by the principles of physics. Additionally, there are factors such as solar insolation, earth’s geology, and volcanic activity. The recent question of anthropogenic contribution to climate change, although controversial, cannot be ignored and needs further research. In this scenario, we will look only at the trend and seasonality from the past century since the time official temperature keeping began. Instead of using absolute temperatures, climatologists use Global Temperature Anomaly for analysis. The term temperature anomaly means a departure from a reference value or long‐term average. A positive anomaly indicates that the observed temperature was warmer than the reference value, while a negative anomaly indicates that the observed temperature was cooler than the reference value. The global time series is produced from the Smith and Reynolds blended land and ocean data set (Smith et al., 2008). This data set consists of monthly average temperature anomalies on a 5° x 5° grid across land and ocean surfaces. These grid boxes are then averaged to provide an average global temperature anomaly. An area‐weighted scheme is used to reflect the MIS480: Predictive Analytics ‐ 3 ‐ reality that the boxes are smaller near the poles and larger near the equator. Global‐average anomalies are calculated on a monthly and annual time scale. The global and hemispheric anomalies are provided with respect to the period 1901‐2000, the 20th century average.1 1. Download the dataset from National Climatic Data Center, National Oceanic and Atmospheric Administration a. Go to https://www.ncdc.noaa.gov/cag/global/time-series to explore the data set. b. Click on Global tab. c. In the Time Series subtab, Choose Timescale — Annual d. Enable Options –Display Trend and per decade e. Click Plot f. You will see a plot of Global Land and Ocean Temperature Anomaly. See Figure 1 Figure 1: Global temperature anomaly chart 1 https://www.ncdc.noaa.gov/monitoring‐references/faq/anomalies.php MIS480: Predictive Analytics ‐ 4 ‐ g. Go to https://www.ncdc.noaa.gov/monitoring-references/faq/anomalies.php h. Choose Anomalies and Index Data tab. i. Choose Time Scale: Monthly, Region: Global, Surface: Land and Ocean, Download CSV (once the data file is displayed in your web browser, make a right-click to save the file). j. Once you are done with downloading the file, then open the file using Excel. We will now prepare this data for Predictive Analysis. i. Delete the first four rows so that only the column headers and data rows remain. ii. Add three columns — Year, Month, Date iii. Use the Excel function Left to populate the Year column referencing the original Year column: type in “=LEFT(A2,4)” to your first cell under Year column. iv. Use the Excel function Right to populate the Month column referencing the original Year column: type in “=RIGHT(A2,2)” to your first cell under Month column. v. Use Excel function Concatenate to populate the Date column to combine Year and Month into the format MM/YYYY: type in “=CONCATENATE(D2, “/”, C2)” to your first cell under Date column. vi. Format the date cells to a date format. vii. Format the year and month cells to number format. viii. Rename the original Year column to Year/Month. Your data should look like Figure 2 Year/Month Value Year Month Date 188001 ‐0.01 1880 01 01/1880 188002 ‐0.13 1880 02 02/1880 188003 ‐0.15 1880 03 03/1880 188004 ‐0.06 1880 04 04/1880 188005 ‐0.09 1880 05 05/1880 188006 ‐0.18 1880 06 06/1880 188007 ‐0.16 1880 07 07/1880 188008 ‐0.09 1880 08 08/1880 188009 ‐0.1 1880 09 09/1880 188010 ‐0.19 1880 10 10/1880 188011 ‐0.28 1880 11 11/1880 188012 ‐0.09 1880 12 12/1880 MIS480: Predictive Analytics ‐ 5 ‐ Figure 2: Splitting Year/Month into Year and Month ix. Save as Temp_Anomaly.xlsx (use Excel file format, not csv) 2. Launch SAP BusinessObject Predictive Analytics 3. Click on Expert Analytics  Expert Analytics 4. Preparing data a. Click on File  New b. Choose Microsoft Excel  Next c. Choose the File Temp_Anomaly.xlsx d. Click Create e. Click on Prepare tab f. Switch to Grid view g. In the Dimension on the left panel, rename the Value column to Temp Anomaly 5. Save the PA file 6. Linear Regression a. Click on Predict tab b. Under Algorithms, double click on R-Linear Regression c. This will connect the Linear Regression algorithm to the Temp Anomaly data source. d. Now Configure Settings for the R-Linear Regression e. Independent Column — Year, Dependent Column –Temp Anomaly. See Figure 3 Figure 3: Building the analysis f. Click Done g. Click Run h. Click OK MIS480: Predictive Analytics ‐ 6 ‐ i. You will see the data grid view of the predicted values of temperature. j. Click on Visualize k. Select Analysis — Analysis 1, Select Component — R-Linear Regression l. Convert the Temp Anomaly and Predicted Values to measures: click cog handle of each dimension & select Create a Measure m. For both measures, change the aggregation of the two to None (from Sum) as follows: n. Choose a Combined column line chart. Y-axis — Temp Anomaly and PredictedValues, Xaxis-Year o. You should see the regression line in Figure 4 Figure 4: Data visualization p. You see the rise in average global temperature anomaly for the past 120 years q. Save MIS480: Predictive Analytics ‐ 7 ‐ r. Save your PA file 7. Triple exponential Smoothing a. Click on Predict tab b. Choose the Temp Anomaly data set from the top left option c. Click Temp Anomaly data icon in the middle panel d. Double-click R-Triple Exponential Smoothing algorithm in the right-side panel e. Click Configure settings f. Enter the settings shown in Figure 5 Figure 5: Conguring settings MIS480: Predictive Analytics ‐ 8 ‐ g. Click Done h. Click Run i. Click OK j. You will see the data grid showing the predicted values of temperature k. Click on Visualize l. Select Analysis — Analysis 1, Select Component — R-Triple Exponential Smoothing m. Convert the Temp Anomaly and Predicted Values to measures n. Change the aggregation of the two to None (from Sum) o. Choose a Combined column line chart. Y-axis — Temp Anomaly and PredictedValues, Xaxis-Year p. You should see the Exponential Smoothing chart in Figure 6 Figure 6: Triple Exponential Smoothing Results MIS480: Predictive Analytics ‐ 9 ‐ q. There are too many data points on this chart. Select a rectangular area around the last two or three years of data r. Then click Filter. You should see a more focused chart of temperatures. s. Save the Visualization. Save the PA file. Question 1: Based on the results, it appears that Triple Exponential Smoothing was able to fit the data well. Why? Question 2: Would single or double exponential smoothing have worked for temperature anomaly analysis? Why? Question 3: Create a forecast of temperature anomaly for the next 24 months using Triple Exponential Smoothing. Create a visualization of those results. See Figure 7 and Figure 8 Hint for Q3: You will need to add one more algorithm to your model (e.g., R-Triple Exponential Smoothing algorithm just like you did in Step 7). This time, however, select Forecast with 24 Periods to Predict in the setting. Figure 7: Temperature anomaly forecast MIS480: Predictive Analytics ‐ 10 ‐ Figure 8: Forecast for next 24 months