This study aims to develop a groundwater potential map using Random Forest (RF) algorithm for the highlands of Somali Regional State. Based on the data availability, 18 groundwater conditioning actors that can significantly affect infiltration rates and groundwater volume were selected and used. Besides, 213 well location and yield data were collected to train and validate the RF model using the ROC-area under curve indices. The multicollinearity test using variance inflation factor and tolerance values showed that only 13 out of the 18 conditioning factors were relevant to develop the RF model for predicting the groundwater potential of SRS. The results in this study also showed a high model accuracy of 92.7%, indicating that the RF model did better at identifying the target groundwater potential map in the area. The groundwater potential map produced by RF model showed that areas with the highest groundwater potential are mainly located in the central Fafan Zone and southern parts of Siti Zone. On the other hand, ‘very low’ and ‘low’ groundwater potentials are located in the northern parts of Siti Zone. In terms of area coverage, ‘moderate’ groundwater potential class covers the largest areas (72.6%), followed by ‘low’ potential (15.2%) and ‘high’ potential (12.1%), whereas very ‘low’ and ‘very high’ classes only cover the tiniest areas. The findings in this study may help policy makers, local managers and hydro-geologists better plan and manage groundwater resources in the Somali Regional State of Ethiopia. Keywords: Machine learning, Groundwater, Siti Zone, Fafan Zone, Conditioning factors