Entropy-Based Hybrid Integration of Random Forest and Support Vector Machine for Landslide Susceptibility Analysis

Round 1

Reviewer 1 Report

The study compares hybrid models generated by integrating a statistical model (Shannon entropy) with two machine learning models (Random Forests and support vector machine), for assessing landslide susceptibility in the Mandi district, Himachal Pradesh, India. Fourteen landslide causative factors were used to model landslide potential: slope gradient, plan curvature, slope aspect, 117 elevation, drainage density, lithology, geology, land use and landcover (LULC), normalized difference vegetation index (NDVI), soil characteristics, lineament density, stream 119 power index (SPI), topographic wetness index (TWI) and distance from the roads.

The study objectives are clearly stated, and the introduction provides important background in the application of algorithms for modeling landslide susceptibility in general. I could not find in the Methods section, the software you used to generate the statistical and machine learning models. Two observations seem important: Highlight the accuracy of models based more on your other parameters (e.g., those derived from the similarity matrix) and less on the AUC of the ROC test. In your discussion, it would be interesting to refer the performance of your proposed hybrid models (e.g., accuracy) as compared with those obtained in previous studies or modeling exercises.

Author Response

 "Please see the attachment." 

Author Response File: Author Response.docx

Reviewer 2 Report

The paper is interesting concerning the integration of the Shannon entropy (SE) statistical model with random forest (FR) and support vector machine (SVM) machine learning models using the optimum feature selection process for landslide susceptibility mapping.

The description of the model parameters is correct.

Flowchart depicting methodology used in the study is clear.

line 312 "It was observed that TWI, drainage density, and NDVI were the primary factors responsible for higher landslide susceptibility in the study area. " How is it possible to prove/show it ?

A general criticism concerns validation and testing. The authors have correctly divided the collection into two parts: training and validation (Figure 1), but are they completely independent? That is, was the whole Figure 2 run up to the last stage on the training set?
and only the final "Model Validation and Comaprison" stage on the validation set?
And how do the results for Figure 5 relate to this run? Is AUC-Prediction calculated from training set?
Please explain how the accuracy metrics were calculated?  e.g. what does AUC stand for?

What "attributes" were extracted from the Field Validation (Landslie Inventory), zero-one, is there a landslide/no landslide?, or somehow corresponding to Landslide Susceptibility Figure4? 

In summary, it is not clear what was compared and was the input for the accuracy analysis: TP (true positive), TN (true negative), FP (false positive), FN (false negative).

This part must be revised in my opinion.

Author Response

 "Please see the attachment." 

Author Response File: Author Response.docx