The existing machine learning based models for personal thermal comfort have traditionally focused on physiological and psychological variations among occupants, and the spatial parameters have been largely overlooked. Field measurements are conducted to collect data and synthesise the collective findings for optimal spatial positioning based on a 24 °C setpoint. The objective of the present study is to investigate and compare the prediction performance made by the machine learning models for personal indoor thermal comfort in air-conditioned office environments using non-spatial parameters (NSP) and spatial parameters (SP). The data was collected from the respondents at four different occupants. A comprehensive data set of NSP and SP is comprised of machine learning models in predicting different thermal comfort situations are Decision Trees (DT), Random Forests (RF), Support Vector Machines (SVM), K-Nearest Neighbours (KNN), Naive Bayes (NB), and Neural Networks (NN). Results indicate a substantial improvement in the accuracy prediction with the Random Forest algorithm outperforming others, enhancing overall accuracy by 38.6 % with spatial parameters for thermal sensation vote (TSV). However, the SVM algorithm improves 50 % accuracy by considering SP input for thermal comfort (TC). Spatial parameters, including the distance between windows and air conditioning units, emerge as critical factors influencing thermal comfort.