Evaluating Deep Learning Based SDAE Prediction Techniques for Predictive Modeling of Landslide Susceptibility

In sensitive locations, landslip susceptibility study helps predict and reduce risks, which leads to better planning and avoidance. The goal of this work is to create an accurate susceptibility model for the Collazzone area in central Italy so that it can look at current risks and predict future landslides. Using a multi-temporal inventory map as the dependent variable, discriminant analysis found 45 environmental parameters. The method includes preprocessing, choosing features, and training the model. Of the four normalisation methods tested in preprocessing-logarithmic, zero-mean, and standard-min-max normalisation was the most accurate. They used MIC and multicollinearity diagnostic to pick features, and the second one worked better at getting rid of extra features. It was new to utilise a Stacked Denoising Autoencoder (SDAE) for training instead of AE and DAE models. With 96.54% accuracy, the SDAE-based framework did better than previous methods. These results suggest that better preprocessing, more reliable feature selection, and more advanced deep learning can make Landslide Susceptibility forecasts more accurate. This will help control hazards, lower risks, and keep communities safe in locations that are prone to landslides.