An LDA–SVM Machine Learning Model for Breast Cancer Classification^†

Round 1

Reviewer 1 Report

The authors proposed a two-step method using LDA feature extraction to boost the performance of SVM on breast cancer. Here are my comments:

 

1. The novelty of the proposed method is not strong. There have been a lot of methods using feature extraction methods to boost machine learning methods’ performance. The authors may want to discuss the novelty of the proposed methods more.

2. Besides the LDA and PCA feature extraction methods the authors used, there are many more feature extraction methods, such as ICA, LLE, t-SNE, and AE. The authors may want to discuss why they choose LDA and PCA or perform more experiments using the other methods and compare their performance.

3. The authors only performed two real data analyses to illustrate the power of their proposed method. They may want to perform some simulation analyses to find more insights into why the proposed method performed well.

4. The performance of the proposed method on the WBCD and WPBC datasets differed a lot. The authors may want to discuss more possible reasons.

 

Author Response

Find attached the responses to your comments

Author Response File: Author Response.docx

Reviewer 2 Report

In this manuscript, Egwom et al. presented a new machine learning model using LDS-SWM to classify breast cancer progression. This work fits the scope of BioMedInformatics journal, and I believe it is of interest to the general audience. In this work, the authors used LDA for feature extraction and then used SVM for classification. This method resulted in high accuracy and recall, and they also showed that using the median value in missing value imputation generated a better performance. I have a few questions about this work:

 

1, The authors used public datasets, including WBCD and WPBC. Please provide more evidence showing that the features used here can be applied easily to other clinical data. Are the measurements included in the regular examination of the patients? Do they need special diagnostic procedures to acquire the data? This is important to show that this model can be re-used and impact the current clinical procedure.

 

2, Please provide more explanation of why using mean or median to handle the missing values. Are those values missing because they were under the detection limit? If so, the authors should use the detection limit (usually a very low number), or impute a minimal value from all the data to impute.

 

3, Please revise figure 1 since a red circle, and a blue line is there. Also, please remove the watermark in Figure 1, which is unprofessional.

 

4, Have the authors tried more advanced classification methods such as random forest or tree-based classification (adaboost, gbm)? Usually, those methods may have a better performance than SVM. Some discussion about other potential methods would be enough since the accuracy, precision, and recall in this study are already very high.

 

Overall, I recognize the value of this work, and I look forward to the revised manuscript.

Author Response

Find attached the responses to your comments

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

No more comments.