Permeability is the ability of a soil to allow liquids to pass through. Of course the soil has a physical characteristic that can be known by laboratory testing. This study aims to determine the physical properties that most affect the coefficient of cohesive soil permeability using the Artificial Neural Network (ANN) tool, the results obtained will later be matched with actual conditions according to the context of engineering geology. The research method begins with an influence or sensitivity analysis using ANN which will produce a correlation coefficient (R). Then, these results will be compared with the influence analysis based on the value of the coefficient of determination (R2). After that, accuracy and error tests will be carried out using the Mean Absolute Percentage Error (MAPE), the highest accuracy values is categorized as the most influential physical property of the 7 physical property parameters, namely liquid limit, plastic limit, plasticity index, %sand, %fines, %silt, and %clay. Based on the result of the analysis, %fines is the parameter that most influences permeability and is able to make very strong predictions with an R value using an ANN of 0.9941875, an R2 value of 0.6336, an accuracy of 99.6962%, and a MAPE of 0.3038%. These results are compared with the existing empirical equations with an accuracy of 96.4393% and MAPE of 3.5607%. It can be concluded that ANN is more effective and optimal in making predictions. In this case, in the context of engineering geology, the more %fines, the smaller the permeability coefficient of the soil.