The transport network is a complex system that benefits from detailed data on user mobility. Analyzing user trajectories through clustering or classification methods can provide valuable insights into mobility patterns. Extracting transport modes from these trajectories using classification methods enhances the understanding of user mobility. The complexity of classification methods varies, with some classifying a few transport modes, such as walking, running, bicycling, and driving. In contrast, others classify up to seven modes or use private, unpublished datasets. A key challenge in transport mode classification is ensuring the comparability of different methods across various contexts. Additionally, comparing results is further complicated by the insufficient use of existing standardized benchmark, which in the case of transport mode classification, must contain a structured testing framework and a dataset on which the testing will be conducted. This research introduces a process for collecting data to develop a new transport mode classification dataset. The goal is to enhance the benchmark by evaluating classification methods across diverse traffic patterns and geographic areas, thereby assessing their spatial independence. Spatial independence is crucial because it ensures that classification methods remain accurate regardless of geographic variations. This improves comparability by enabling consistent evaluation of methods across regions, as the improved benchmark addresses spatial independence and ensures robustness for real-world deployment. The current benchmark in literature examines three types of independence: user, position, and time independence. Our tests employ a multilevel method based on Transition State Matrices (TSMs) and the Random Forest (RF) algorithm for transport mode classification. The results demonstrate that the multilevel method maintains spatial independence and achieves higher accuracy compared to the original benchmark problem.