Read Journals 1795 – Vol 68 No 1  – Prediction of operating speed on horizontal curve and tangent: A comparative study of multiple linear and principal component regression

Operating speed is considered a crucial factor in evaluating geometric design consistency on highways. The operating speed models developed in earlier studies predominantly employed the approach of multiple regression analysis. Using this approach, it is possible to find the presence of multicollinearity among independent variables in the developed models. To examine the effect of multicollinearity, the current research formulated prediction models for operating speed on both horizontal curves and tangent sections using multiple linear regression (MLR) and principal component regression (PCR). The MLR analysis is applied to model the operating speed, and it showed that multicollinearity (variance inflation factor > 4) exists among various independent variables, which affects the predictability of the model developed using MLR. Therefore, principal component analysis (PCA) was applied to eliminate the multicollinearity between independent variables. The PCA results revealed that only the first two components (PC1 and PC2) with eigenvalues greater than one were selected for PCR analysis. Furthermore, when comparing the results obtained by MLR and PCR, the PCR model yields good predictions for both curved and tangent section cases. Finally, the PCR model was selected as the best-fitted model for predicting operating speed on the curve and tangent sections, as it had the lowest variable inflation factor values and the highest R2 values compared with the MLR model. According to the findings, this study suggests employing PCR instead of MLR when multicollinearity is present in the dataset.