ANN classification for geotechnical stability

This research presents an in-depth exploration of the challenges and benefits of using a neural network model for a complex classification task. Central to the investigation was the process of hyperparameter tuning, which, while ensuring model optimization, also introduced significant computational demands. A strategic transformation from multi-class to binary classification was made, resulting in increased accuracy but limited prediction specificity. The study revealed that while deep learning models demonstrated superior performance in terms of accuracy, their intrinsic complexity posed challenges related to transparency and interpretability. Comparisons between traditional machine learning models and deep learning models illustrated the trade- offs between prediction time and accuracy. Furthermore, a user-friendly application interface was designed and developed using Tkinter, emphasizing intuitive interaction and visual feed- back mechanisms. The deep learning model achieved average accuracies of 95.24% and 98.81% for three-class and binary classifications, respectively. Findings from this study under- score the promise of deep learning in classification tasks, highlighting areas for future research in optimization, interpretability, and user-friendly design.