•  
  •  
 

Abstract

Walkability assessments are becoming more popular, as walking offers numerous health, environmental, and economic benefits to communities. However, previous studies on ferry terminal walkability assessment have been inadequate. This study aimed to develop a wearable sensor system to automatically assess walkability at ferry terminals without conducting surveys. We applied seven machine learning (ML) classifiers to detect different walking environments, including flat ground (FG), downhill slope (DS), uphill slope (US), and uneven surface (UE). The ML models were evaluated across different combinations of classes: 2-class (FG vs. UE), 3-class (U) (FG vs. US vs. UE), 3-class (D) (FG vs. DS vs. UE), and 4-class (FG vs. DS vs. US vs. UE). Among these, support vector machine (SVM) classifiers had the best area under the receiver operating characteristic curves (AUCs) for the 2-class, 3-class (U), and 4-class datasets with 0.970, 0.920, and 0.922, respectively. AdaBoost (AB) performed the best in 3-class (D) with an AUC of 0.953. The least absolute shrinkage and selection operator exhibited better performance in classifying walking environments than maximum relevance and minimum redundancy. This study assessed passenger walkability and improved the built environments at ferry terminals by identifying uncomfortable walking conditions. Furthermore, the results contribute to the development of a passenger walkability evaluation system utilizing intelligent sensors and to the economic revitalization of communities near ferry terminals.

Download
COinS