<p>The evaluation of the thermal perception of urbanities in outdoor spaces requires the application of several interviews usually gathering basic information regarding human thermal comfort. This research applies the Artificial Neural Networks (ANN) technique to predict outdoor thermal sensation, due to its generalization capability, handling high-dimensional data, and solving nonlinearities problems, offering better predictions compared to typical parametric approach. The focus is to evaluate the ability of the ANN models to estimate thermal sensation votes (TSV) declared by pedestrians in outdoor spaces in a tropical climate, considering basic and complementary factors associated with thermal comfort. In the first stage, ANN models included meteorological, individual, and anthropometric predictor variables (a database of 685 interviews). In the second stage, additional features associated with the interview location, emotional state, and declaration of air condition use were applied. The best TSV prediction was achieved with an ANN model comprising four hidden layers and two neurons (prediction error = 0.153 on the TSV scale), considering all the basic factors investigated the first stage. The model prediction performance was improved by adding the feature related to the interview place (prediction error = 0.115). In contrast, including the emotional state significantly worsened the accuracy of the ANN model (prediction error = 1.146), while air conditioning use slightly worsened it (prediction error = 0.245) despite improving the person coefficient. The findings suggest that features related to the interview location and air conditioning acclimatization should be included in outdoor thermal comfort questionnaires to enhance the ANN's predictive performance regarding thermal perception.</p>