This study aims to highlight the performance of the Support Vector Machine (SVM) and the Random Forest (RF) Machine Learning (ML) algorithms to evaluate the changes in Land Use and Land Cover (LULC) in one of the watersheds belonging to a semi-arid environment in the Northwest of Tunisia (Tessa Watershed), between the years 1993 and 2023. Remote sensing (RS), statistical analysis, and geographic information systems (GIS) are employed in tandem in the present research. Accuracy metrics enable the evaluation of image classification methods by calculating producer's accuracy, user's accuracy, overall accuracy, and the Kappa coefficient. In satellite image classification, Machnie Learing ML algorithms are increasingly used over classic classification methods. In terms of efficiency, the SVM and the RF methods, rank highest among these algorithms. The SVM method produced an overall accuracy of 95.91% with a Kappa value of 0.867 on the 1993 Landsat 5 TM image, whereas the 2023 Landsat 8 OLI image achieved 88.79% accuracy with a Kappa value of 0.86. In comparison, the RF algorithm achieved an overall accuracy of 86% with a Kappa value of 0.8 for the 1993 image and an overall accuracy 81% with a Kappa value of 0.77 for the 2023 image. The transition matrix allowed the identification of spatiotemporal LULC changes. The results of the current study which considers six LULC classes, show that the outputs of the SVM classifier are more in line with data that is derived from ground truth compared to the RF classifier results.