Assessing the Number of Criteria in GIS-Based Multicriteria Evaluation: A Machine Learning Approach

Abstract

The analytical hierarchy process (AHP) is a widely used approach and a decision rule to derive criteria weights in geographic information system-based multi-criteria evaluation (GIS-MCE). However, one limitation of the AHP method is that it constrains the number of criteria that can be meaningfully weighted to typically seven to nine criteria. Recently, machine learning (ML) techniques have emerged as a compelling alternative for deriving criteria weights. This research aims to assess the capabilities of ML-MCE in handling a larger number of criteria and is specifically applied to a case study of urban suitability analysis. The random forest (RF) ML technique is used to evaluate the ability of the MCE method to handle up to 27 criteria. Geospatial data from the Metro Vancouver Region, Canada, are used, with the criteria subdivided into 11 groups starting with the most basic seven criteria and incrementally adding two new criteria per group. The results indicate the RF-ML approach can manage a larger number of criteria compared to the traditional AHP approach, with 15 criteria providing a meaningful upper threshold, demonstrating its potential to accommodate a wider range of stakeholder preferences for complex urban suitability analysis contexts.