Efficient local AADT estimation via SCAD variable selection based on regression models

Abstract

In transportation networks, Annual Average Daily Traffic (AADT) estimation is very important to decision making, planning, air quality analysis, etc. Regression method may be the most popular one for estimating AADT on non-counters roads. Most literatures focus on how to collect different groups of predicting variables, and select significant variables by t-test and F-test. However, there is no theory on the validity of these multiple selecting steps. Furthermore, variables they collected for high functional class roads may be not suitable for the estimation of local AADT because of lacking counters. This paper focuses on the estimation and variable selection for the local AADT using different groups of variables. The variable selection by smoothly clipped absolute deviation penalty (SCAD) procedure is proposed. It can select significant variables and estimate unknown regression coefficients simultaneously at one step. The estimation algorithm and the tuning parameters selection are presented. The used data is from Mecklenburg County of North Carolina in 2007 for demonstration. The proposed method shows that our selection procedure is valid and it further improves the local AADT estimation by incorporating satellite information. The proposed method outperforms some other regression method when it is applied to local AADT estimation.