Urban accessibility is one of the main factors that determine the sustainability of an urban area. The Urban Accessibility Index (UAI) needs to be estimated accurately for transportation planning and policy decision-making. This work presents a wide-ranging machine learning-based framework that uses basic urban and transport network features to predict Urban Accessibility Index (UAI). The features or predictors that were identified are Nodes, Links, Road Network Length, Built-up Area, and Area; these features demonstrate the structural properties of the urban system. Ten various types of models were tested and compared using the dataset presented; these were Linear Regression, Ridge Regression, Lasso Regression, Decision Tree, Random Forest, Extra Trees, Gradient Boosting, Support Vector Regression, K-Nearest Neighbours, and finally the XGB-Booster model. The models were evaluated in the form of the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2). From the comparative study, it emerges that the ensemble and instance-based techniques always guarantee better results than their linear and single-tree competitors. K-Nearest Neighbours model, out of all algorithms, received the highest prediction accuracy by way of the lowest testing RMSE of 2.67 and the highest R² value of 0.87, with just a little less from Random Forest and Extra Trees models. Gradient Boosting and XG-Boost, though showing almost perfect training accuracy, show a decrease in testing performance, showing a case of overfitting. In summary, the results have proved the ability of machine learning techniques to recognize the nonlinear relationships that influence the accessibility in cities. The introduced model of measurement of accessibility is a dependable and easily expandable instrument, thus assisting planners in the urban transport sector who are both sustainability-oriented and data-driven.
Kurre et al. (Wed,) studied this question.