Forests are vital to Earth's ecology; however, refugee settlements like the densely populated Jewi Refugee Camp, Gambella region, contribute to forests resource depletion. To support mitigation effort, accurately assessing forest cover change and fragmentation process using optimal remote sensing models and datasets suited to the specific location is essential for implementing empirically informed management actions. Therefore, this study assesses the impact of refugee settlements on forest resources in Jewi Kebele. To address this objective, the study computed three machine learning classifiers: Random Forest (RF), Classification and Regression Trees (CART), and Support Vector Machine (SVM), on two input dataset category: Sentinel-2 multispectral bands alone, and a composite dataset that combined Sentinel-2 multispectral bands with their derived indices and Sentinel-1 GRD bands enhanced by Gray Level Co-occurrence Matrix (GLCM) textural features. The classifier and dataset combination that yielded the highest accuracy was then used to quantify forest cover change and fragmentation process. Forest fragmentation was analyzed using the Landscape Fragmentation Tool (LFT), which classifies forest areas as core, perforated, patch, and edge zones. To examine the impact of fragmentation on forest species diversity, alpha and beta wood species diversity indices were applied across fragmentation zones, and 95 samples were collected through stratified random sampling across these zones, using 20 m × 20 m plots within a 2.5 km × 2.5 km grid. The classification accuracy results showed that in 2024, RF, CART, and SVM classifiers achieved Kappa coefficients of 0.904, 0.850, and 0.392 on Sentinel-2 multispectral bands, and 0.931, 0.827, and 0.710 on the combined dataset, respectively. The combined dataset’s achieved higher accuracy and prompted its use for 2015 and 2020, performed Kappa coefficients for RF of 0.938 (2015), and 0.937 (2020); CART of 0.894, and 0.858; and SVM of 0.874, and 0.592. The outperforming RF with the combined dataset result revealed, forest cover declines of 5.48% from 2015 to 2020 and 14.92% from 2020 to 2024. Also, the forest fragmentation result showed, reduced continuous (core) forest by 1.1% (2015–2020) and 42.5% (2020–2024), perforated forest by 9.55% and 8.38%, and edge forest by 28.9% and 22.4%, shifting to non-forest and other zones, while isolated (patch-zone) forest increased by 21% and 13.3%. Besides, across the current remaining fragmentation zones, alpha diversity indices revealed species diversity patterns, continuous forest showing high diversity (Shannon: 2.5, Simpson’s: 0.915, Pielou’s: 0.976), followed by perforated forest (2.38, 0.88, 0.928), while isolated forest (2.11, 0.86, 0.91) and edge forest (1.47, 0.70, 0.76) had lower diversity. Also, beta diversity indices revealed connectivity of each zone with continuous forest through shared species, with perforated forest most connected (Jaccard similarity: 42%, Morisita-Horn index: 78%), followed by isolated forest (34.8%, 52%), and edge forest least connected (3.5%, 43%). These findings indicating that forest cover change has led to fragmentation by refugee settlements, which has adversely affected the species diversity of forests. These results underscore the need for targeted restoration to strengthen ecological resilien