Global Navigation Satellite Systems (GNSS) support numerous applications, including mission-critical ones that require a high level of integrity for safe operations, such as air, maritime, and land-based navigation. Fault Detection and Exclusion (FDE) is crucial for mission-critical applications, as faulty measurements significantly impact system integrity. FDE can be applied within the positioning algorithm in the measurement's domain and the integrity monitoring domain. Previous research has utilized a limited number of Machine Learning (ML) models and Quality Indicators (QIs) for the FDE process in the measurement domain. It has not evaluated the pseudorange measurement fault thresholds that need to be detected. In addition, ML models were mainly evaluated based on accuracy, which alone does not provide a comprehensive evaluation. This paper introduces a comprehensive framework for traditional ML-based FDE prediction models in the measurement domain for pseudorange in complex environments. For the first time, this study evaluates the fault detection thresholds across 40 values, ranging from 1 to 40 m, using six ML models for FDE. These models include Decision Tree, K-Nearest Neighbors (KNN), Discriminant, Logistic, Neural Network, and Trees (Boosted, Bagged, and Rusboosted). The models are comprehensively assessed based on four key aspects: accuracy, probability of misdetection, probability of fault detection, and the percentage of excluded data. The results show that ML models can provide a high level of performance in the FDE process, exceeding 95% accuracy when the fault threshold is equal to or greater than 4 m, with KNN providing the highest FDE performance.