Diabetes has always been the focus of attention globally, with its high impact on mortality. Over the years, machine learning and ensemble learning have been found to be potential investigations to conduct an early prediction of diabetes. Due to rapid development in these fields, it is necessary to provide a comprehensive review with overview information for early detection studies of diabetes, particularly in the employment of ensemble learning that has the benefit of merging multiple algorithms. However, most of the previous review studies were less in-depth and not mainly focused on ensemble learning applications. In this paper, a systematic review study in the area of ensemble learning for diabetes prediction is presented to overcome this issue, which included a total of 98 studies published between 2014 and 2024. Based on the methodologies, data extraction, and study processes, the result of key findings especially appraises the current state of knowledge, such as highlighting the trends in the case study, the critical review of each ensemble technique, and the advantages as well as the disadvantages of ensemble learning. Additionally, the identification of limitations is revealed in the tasks of the dataset, and the analysed studies also confirmed the opportunities for future work are in the directions of data sources, ensemble deep learning, and data preprocessing. Thus, the results of this work aim to provide a better understanding of the field area with major findings and new insights for further expansion in this scope of ensemble learning in diabetes early detection.