Prediction of landslide tsunami run-up through ann-based models

New prediction models based on multilayer perceptron are proposed which successfully predict the maximum run-up of landslide-generated tsunami waves and the roles of parameters affecting the maximum run-up are assessed. The input to the models consists of approximately 55,000 rows of data employing six predictors, namely cross section of the landslide profile, the beach slope angle, the initial slide submergence, the vertical thickness and the horizontal length of the sliding mass, and the time of the maximum run-up. The target variable is the maximum tsunami run-up. An optimization study was first performed and a subset of 9,000 randomly sampled rows was determined to represent the whole data. A total number of 63 models consisting all possible combinations of the six parameters were then constructed for the 9,000-row data set, in an attempt to predict the maximum run-up. The prediction capability of the models was assessed by calculating Root Mean Square Error, Mean Absolute Error, Mean Absolute Percentage Error (MAPE), and Multiple Correlation Coefficient (R) as performance metrics. The MLP-based models led predictions with a minimum MAPE of approximately 1.1% and with R=1, revealing that the slide thickness has the largest impact on the maximum tsunami run-up, whereas the slide cross-section and the slope angle have minimal effect. Further, a backward elimination algorithm equipped with ReliefF feature importance method is utilized as a practical tool. Comparison with existing literature showed the reliability and applicability of the offered models. This approach can be therefore used as a fast and accurate methodology to support or when there is lack of analytical or numerical modeling.