A comparison of statistical and neural network models for forecasting educational spending
Bruce D. Baker, Craig E. Richards · 1997
The objective of the study is to determine the comparative effectiveness of statistical and neural network models for forecasting nationally aggregated educational spending. Two neural architectures, Backpropagation and Generalized Regression Neural Networks have been suggested for use in time series analysis. In this study, the predictive accuracy of Backpropagation and GRNN models is compared with the multiple linear regression model used by the National Center for Education Statistics for projecting educational spending. Input variables for all neural network models are the same as those used by William Hussar and Deborah Gerald in the NCES report Projections of Education Statistics to 2005. Exploratory analyses are also conducted with state level data. Neural networks are expected to outperform regression models on the premise that educational expenditures are just one component of a complex economic system and that complex systems require higher levels of behavioral analysis. It is believed that such analyses cannot adequately be modeled with linear equations. The results of the study confirm these expectations.