A Combined Method to Short Term Demand Forecast Accuracy: Data Mining Using A Bottom-Up Approach
College of Computing, Engineering & Construction
Master of Science in Computer and Information Sciences (MS)
Dr. Robert F. Roggio
Dr. Sanjay P. Ahuja
Dr. Adel El-Ansary
Dr. Judith L. Solano
Dr. Peter A. Braza
For a transportation enterprise, accuracy of a short term demand forecast can dramatically affect shipping network utilization, sales revenue, operational expenses, resource allocation, and strategic operational planning. The aim of this thesis is to investigate the performance characteristics of a combined bottom-up approach to short term demand forecasting. The bottom-up approach to this investigation involves grouping the business unit data into three aggregation levels; origin-destination (OD) pairs, origination city, and origination state. Forecast values for each of these aggregation levels are then rolled-up into the overall business unit forecast. The goal is to spread the error across individual forecasts, which may lead to an improved overall business unit forecast. Another benefit of grouping the data into segments is to be able to forecast at a more localized level, so operations may better utilize the shipping network and apply resources where needed. The forecast values are generated from historical data analyzed by various data mining models. Data mining techniques, such as Artificial Neural Networks (ANN) and time series analysis (via Microsoft SQL Server ARTxp algorithm), are especially valuable in recognizing patterns in historical data of large datasets. Forecasting performance of each of these data mining models is quantified by standard statistical measures such as the mean absolute deviation (MAD), mean square error (MSE), and mean absolute percentage error (MAPE).
Roberts, Brian J., "A Combined Method to Short Term Demand Forecast Accuracy: Data Mining Using A Bottom-Up Approach" (2009). UNF Graduate Theses and Dissertations. 1059.