0
$\begingroup$

I am in a non-computer science field, and machine learning is being blatantly misused in my field.

I recently got a journal paper to review, where the researchers used machine learning to develop a predictive model of an experimental result. I think that the dataset they have utilized in their work is too little for successful feature selection and for determination of the important features.

I would like to get an insight into whether the researchers have done a sound study or not?

The dataset used in the study

In the figure, S1...S8 are 8 different samples. According to the researchers, Machine learning can be used to predict the values of A,B,C... for an unknown sample.

For training, they have used 6 samples, and 2 for testing. Now they have claimed in the paper that they have used 28 cases for feature analysis, which they obtained by using the 6 data and implementing extra trees, correlation analysis and SelectKbest analysis.

I don't understand, how they produced the 28 datasets. And they have later validated the model using these 28 cases.

I don't know how to review this work. I feel that it is not well-researched work, but I wanted some expert opinion before putting forward my thoughts. I might be absolutely wrong with this.

The researchers have claimed 94% predictive accuracy using the ANN model.

$\endgroup$
3
  • 1
    $\begingroup$ How to adequately model a statistical phenomenon obviously depends on the distributional profile of the phenomenon in question. This is not a programming problem. I have nominated this for migration to our sister site Cross Validated but I suspect they will only tell you the same thing. $\endgroup$
    – tripleee
    Dec 29, 2019 at 16:46
  • $\begingroup$ They probably got 28 datasets by using k-fold cross-validation. $\endgroup$
    – Mark Meyer
    Dec 29, 2019 at 16:46
  • $\begingroup$ @MarkMeyer But is 6 datasets enough to do k-fold cross-validation? $\endgroup$
    – rcty
    Dec 29, 2019 at 16:56

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.