See the attached file with the comments. 

This paper presents a very innovative and relevant study, showing the possibility of using ML techniques to predict the type of aerosols. The work is very well written and structured. However, the following points need to be better detailed.

Line 135: Why was the median used to fill in the gaps? Did you try to use other techniques? Perhaps the use of machine learning techniques could generate a more robust filling.

Table 1: Why were these groups of hyperparameters exclusively selected? Was any analysis of the importance of hyperparameters performed? This can severely affect the final performance of the models, especially for neural networks.

Figure 4: I recommend increasing the font size of the axes.

Figure 4: How did you deal with the problem of the imbalance of the dataset? Because "continental polluted" tends to have worse performance due to the smaller number of data.

Line 285: I expected better results from the NN model with depolarization data, since you have more information about the particle analyzed. Isn't this difference associated with the data input format in the model? Was any preprocessing performed to normalize them?

Figure 5: Considering the use by other users, I think it is important to comment on the computational cost of each model.

Section 3.2.3: Was an analysis of multicollinearity between the features performed? This can affect the importance of each one in the model, as well as the performance of the final model.

Line 323: Because of this statement, I expected that depolarization would present better results in the MLP Classifier.

Line 363: I recommend reviewing the imbalanced dataset issue because if this is not corrected, the cases that are less present in the training tend to perform worse.