Rangeland degradation and bush encroachment pose a major risk to food security, the preservation of biodiversity, ecosystem services, and livelihoods. A major challenge in semi-arid regions is the dearth of information showing land degradation in nature reserves. This study examines land cover/land use (LCLU) changes in Mokala National Park (MoNP), South Africa, focusing on bush encroachment and rangeland degradation. This study evaluates the performance of three machine learning algorithms (k-nearest neighbour (kNN), random forest (RF), and support vector machine (SVM)) in classifying LCLU using Landsat satellite imageries from 1988 to 2023. The selection of these algorithms was based on their proven track record of robustness and consistency over extensive temporal datasets. These algorithms were chosen for their computational efficiency and interpretability, making them suitable for handling the satellite imagery data spanning from 1988 to 2023. Additionally, they provide a reliable baseline for future research to compare and benchmark against more advanced methods. The study aims to improve the accuracy of LCLU classification by comparing the effectiveness of these machine learning algorithms. Training and testing samples for LCLU classification were collected through expert visual interpretation of historical Google Earth satellite imagery and corresponding high-resolution panchromatic images from the Landsat series. The study implemented a comprehensive evaluation framework, incorporating metrics such as overall accuracy, kappa coefficient, and class-specific accuracy to assess the performance of each algorithm. Additionally, the study analysed the temporal stability and adaptability of these algorithms to different periods and varying landscape dynamics. The results show that the RF algorithm has the highest overall accuracy values, ranging from 85% to 95%. Satellite images show an increase in bush encroachment at the expense of grasslands, with woody plants increasing from 4786 ha (1988) to 15,256 ha (2023). The study demonstrates that the RF algorithm significantly outperformed both kNN and SVM in accuracy, while also highlighting a notable increase in bush encroachment, with woody plants increasing substantially from 1988 to 2023. Overall, this information is important for park officials, conservation planners, and other stakeholders to develop ways to address land degradation to protect the resilience and ecological balance of arid areas that are increasingly in danger from bush encroachment. This study provides valuable guidance for selecting appropriate machine learning algorithms for LCLU classification in remote sensing applications. The conclusions of the study contribute to the broader field of Earth observation by highlighting the strengths and limitations of each algorithm, thus informing future research and practical implementations of LCLU analysis.