Climate has played a significant role in shaping the distribution of mammal species across the world. Mammal community composition can therefore be used for inferring modern and past climatic conditions. Here, we develop a novel approach for bioclimatic inference using machine learning (ML) algorithms, which allows for accurate prediction of a set of climate variables based on the composition of the faunal community. The automated dataset construction process aggregates bioclimatic variables with modern species distribution maps, and includes multiple taxonomic ranks as explanatory variables for the predictions. This yields a large dataset that can be used to produce highly accurate predictions. Various ML algorithms that perform regression have been examined. To account for spatial dependence in our data, we employed a geographical block validation approach for model validation and selection. The random forest (RF) outperformed the other evaluated algorithms. Ultimately, we used unseen modern mammal surveys to assess the high predictive performances and extrapolation abilities achieved by our trained models. This contribution introduces a framework and methodology to construct models for developing models based on neo-ecological data, which could be utilized for paleoclimate applications in the future. The study aimed to satisfy specific criteria for interpreting both modern and paleo faunal assemblages, including the ability to generate reliable climate predictions from faunal lists with varying taxonomic resolutions, without the need for published wildlife inventory data from the study area. This method demonstrates the versatility of ML techniques in climate modeling and highlights their promising potential for applications in the fields of archaeology and paleontology.