Abstract: A method includes accessing a first instance of a first web application screen and a second instance of the first web application screen, generating key-value pairs for each web element identified in the first instance and the second instance, and generating a first element map based on the key-value pairs. The method further includes accessing a third instance of the first web application screen, and generating an updated element map based on the first element map and a second element map generated based on the third instance. The method includes executing a machine-learning model trained to generate a prediction of key-value pairs for instances of a second web application screen based on the updated element map and a wireframe of the second web application screen, and generating a final element map based on the prediction of key-value pairs and the wireframe of the second web application screen.