Whereas engineering applications require orthometric heights of points, the Global Navigation Satellite System (GNSS) offers ellipsoidal heights. To determine the orthometric heights of points, levelling measurements must be made, which is a tedious and drawn-out procedure. In this study, an artificial neural network (ANN) was used to integrate the best global geoid model in this region with the original geoid model, resulting in a local geoid model with high accuracy. The accuracy of five global geoid models (GGMs) EGM2008, GECO, XGM2019e_2159, EIGEN-6C4, and SGG-UGM-1, were tested in this research. The geoid height accuracy in the study area, as determined by EGM2008, has an RMSE of around 0.20 m. While geoid height accuracy was calculated from EIGEN-6C4, GECO and SGG-UGM-1 has an RMSE of about 0.15m ,0.15m and 0.18m, respectively. The results showed that, XGM2019e_2159 is the most commonly used global model for geoid surface modelling in the Mediterranean Sea, with a standard deviation of 14 cm. It worth noted that, there has been a significant improvement in results with ANN created local geoid models. When creating the initial local geoid models in the study area, the ANN model accuracy ranged from 0.07 to 0.042 m. However, the local geoid model created by integration between the control points was done using ANN and the XGM2019e_2159 global model is about 60% more accurate than models created with the GNSS/leveling points only, and this becomes apparent as the distance between the control points increases.