Novel Deep Learning Model Optimized by Random Search or Grid Search Method for Soil Erosion Susceptibility Prediction

Abstract

Soil erosion is the process by which soil particles are removed from the Earth's surface. There are three stages to soil erosion: displacement, migration, and deposition. The rate of soil erosion is influenced by various factors, including infiltration, soil type, soil structure, and land cover. Soil erosion causes soil deposition in some areas and soil loss in others. One of the primary issues in meteorology is the prediction of soil erosion. Numerous methods for forecasting precipitation have been put forth, drawing from deep learning, machine learning, and statistical analysis approaches. In this paper, we compare between the Random Search and Grid Search optimization which combine with CNN, RNN, LSTM and GRU algorithm (GS_CNN, GS_RNN, GS_LSTM, GS_GRU, RS_CNN, RS_RNN, RS_LSTM, RS_GRU,) for soil erosion prediction. These models facilitate planning for soil preservation and land management techniques by improving our understanding of and capacity to forecast the dynamics of soil erosion. There are 236 instances with 11 features in the dataset that was used for this study. Six evaluation metrics were computed: accuracy, precision, recall, F1 score, Matthew’s correlation coefficient (MCC), and Area Under the Receiver Operating Characteristic Curve (AUC) to assess the efficacy of the employed classification technique. With an accuracy of 98.592%, the CNN, GS_CNN, GS_RNN, RS_CNN and RS_RNN models outperformed other machine learning methods and earlier research on the same dataset, according to the experimental results.