GLINA ARTS A FORECAST ON FALL SESSION ENROLLMENT ALA AL-LOZI ANTHONY ALEXANDER MATTHEW COKER TOREY HERZOG PROBLEM Apply statistical techniques to the Glina Arts data provided to predict demand for courses and course media types for Fall2016 Agenda 1. How the problem was approached 2. What was considered 3. Final 2016 Fall enrollment prediction 4. Recommendations 5. Conclusions APPROACH Step One: Decide what forecasting method to predict the demand of Fall session 2016 Exponential Triple Smoothing Time Series – Registration Date Demand – Fall Session Enrollment APPROACH Step Two: Evaluated how registration date connected to fall session enrollment • Adult Dataset: August 15, 2014 to November 18, 2014 • Camps Only Dataset: August 18, 2014 to November 21, 2014 • Conclusion: Registration from 3rd week of August to 3rd week in November indicates count of Fall Session enrollees APPROACH Step Three: Create a pivot table to obtain a total number of individuals enrolled in each course media type Organize pivot table were a count of enrollees per week is indicated by course media type APPROACH Step Four: Enter the enrollee count to its corresponding week in a new dataset that includes every week of the original data range to each course media type of Adult Classes and Camps Only APPROACH Step Five: Run a Exponential Triple Smoothing Forecast to each new course media type worksheets from Adults Classes and Camps Only in Excel 2016 WHAT WAS CONSIDERED Models Linear Regression Would not have given the total number of people registered by course or course media type Exponential Single Smoothing Suitable for forecasting data with no trend or seasonal pattern Exponential Triple Smoothing Time series forecast – appropriate for dataset Considers Seasonality Easy to use one Excel 2016 TRIPLE EXPONENTIAL SMOOTHING This method is comprised of several statistics that are required for accurate forecasting: Alpha Data Smoothing Factor Desired Level: ≤0.05 Probability of predicting Type I error Beta Trend Smoothing Factor Desired Level: ≤0.20 Probability of predicting Type II error Gamma Seasonal change smoothing factor Ordinal data correlation TRIPLE EXPONENTIAL SMOOTHING This method is comprised of several statistics that are required for accurate forecasting: Mean Absolute Scaled Error (MASE) Measures the accuracy of the forecasts Symmetric Mean Absolute Percentage Error (SMAPE) Measures the accuracy based on the error percentage Mean Absolute Error (MAE) Measures how close the forecasts are to the eventual outcomes Root Mean Square Error (RMSE) Measures the average of the square roots of the errors and deviations WHAT WAS CONSIDERED Demand Course Only small numbers of people enrolled by course - no trend would have been seen using ETS by course Course Media Type More appropriate way to see a trend compared to evaluating each course 8/18/2013 9/22/2013 10/27/2013 12/1/2013 1/5/2014 2/9/2014 3/16/2014 4/20/2014 5/25/2014 6/29/2014 8/3/2014 9/7/2014 10/12/2014 11/16/2014 12/21/2014 1/25/2015 3/1/2015 4/5/2015 5/10/2015 6/14/2015 7/19/2015 8/23/2015 9/27/2015 11/1/2015 12/6/2015 1/10/2016 2/14/2016 3/20/2016 4/24/2016 5/29/2016 7/3/2016 8/7/2016 9/11/2016 10/16/2016 11/20/2016 FINAL PREDICATION Adult Classes – Woodturning 25 20 15 10 5 0 -5 -10 -15 -20 -25 Registration Count Forecast(Registration Count) Lower Confidence Bound(Registration Count) Upper Confidence Bound(Registration Count) 8/18/2013 9/22/2013 10/27/2013 12/1/2013 1/5/2014 2/9/2014 3/16/2014 4/20/2014 5/25/2014 6/29/2014 8/3/2014 9/7/2014 10/12/2014 11/16/2014 12/21/2014 1/25/2015 3/1/2015 4/5/2015 5/10/2015 6/14/2015 7/19/2015 8/23/2015 9/27/2015 11/1/2015 12/6/2015 1/10/2016 2/14/2016 3/20/2016 4/24/2016 5/29/2016 7/3/2016 8/7/2016 9/11/2016 10/16/2016 11/20/2016 FINAL PREDICATION Camps Only – Metals 30 20 10 0 -10 -20 -30 Registration Count Forecast(Registration Count) Lower Confidence Bound(Registration Count) Upper Confidence Bound(Registration Count) FINAL PREDICATION FINAL PREDICATION FINAL PREDICATION CONCLUSION Adults Classes No courses had an Alpha ≤ 0.05 Camps Only “Camps” is the only course with an Alpha ≤ 0.05 The data is not very reliable. It is difficult to make an accurate prediction. RECOMMENDATIONS Provide Current Data Current data will allow a more accurate prediction Dataset is of low quality Significant cleaning was made in past data project. Can conclude there are possible data entry errors that may have effected the results of the forecasting analysis Implement more stringent governance of data Train data collectors on how to properly gather clean data Consider creating a locked Excel document with dropdowns Clarify “No Season Specified” entry under “Season” Majority of “No Season Specified” entries fell under the second week of August 2013 to the third week of November 2013 Entries appear to be Fall Session data, but the group felt uncomfortable making this assumption QUESTIONS