Understanding Travel Mode Choices (TMCs) and the conditions influencing these decisions is crucial for promoting sustainable transportation, including the use of public transport. One method to explain TMCs is through machine learning models trained with real trip data. However, a promising direction involves combining the representation of model outputs in terms of the probability or odds of selecting a particular mode of transport with the presentation of these values as a function of geographic location to reveal the variability of mode choices in different city areas. Thus, in this work we propose a method providing spatial distribution of predicted travel mode choices. The proposed method takes advantage of two categories of trip datasets: real-world data and synthetic data. The real-world survey data are used to train a classification model predicting travel mode choices, which is subsequently employed to predict probabilities for the trips in a much larger synthetic dataset ensuring higher spatial density of trips. We compare the use of machine learning and logistic regression for the analysis of the spatial distribution of the likelihood of choosing a particular mode of transport. Importantly, we consider exact trip end point coordinates rather than zone centroids and consider time-dependent rather than static level-of-service attributes. The analysis reveals the spatial variability of choices caused by e.g. time-dependent walking distances needed to use public transport.