I am currently working on a database and I have tried various methods of machine learning and deep learning, all leads to poor classification/prediction. However, this might show that the two groups have no significant difference. For example, the database I have is a collection of audio speech and their associated depression level. I am assuming that the culture of where I am from and the strong religious believe, prevented the society from going too deep into depression. However, I need to find some literature to back up this statement. But, it seems that there are no significant difference between the acoustics of depressed and non depressed. Can I write these poor results in a paper? Any tips?
Thank you
Hello Dr Nik, Yes sure you can write a paper and publish it in Scopus journals...
I can present some tips for you:
1- You can change the Chunk size of the voice sample and then check the results again (Usually when the chunk size (sec) is big, the results will increase (accuracy). While when the Chunk size is small, it sounds like a challenge to get high results.
2- Also, when the databases are equal (e.g., depressed and non depressed databases), the results often become higher
2- Give some significant points in the conclusion section in terms of this area.
3- Write the paper in very good English.
4- Moreover, check out these articles may help you in this area:
https://www.sciencedirect.com/science/article/pii/S153204641830090X
Thesis ''Automatic assessment of depression from speech: paralinguistic analysis, modelling and machine learning''
https://www.sciencedaily.com/releases/2019/07/190712133338.htm
https://ieeexplore.ieee.org/abstract/document/9077698
And all the best (:
Salam Dr.Nik
I am Nouar. Do you remember me?
I have very deep experience in machine and deep learning. I think i can help you to have better results if you share me your data.
This is my email: [email protected]
Hello Dr Nik, Yes sure you can write a paper and publish it in Scopus journals...
I can present some tips for you:
1- You can change the Chunk size of the voice sample and then check the results again (Usually when the chunk size (sec) is big, the results will increase (accuracy). While when the Chunk size is small, it sounds like a challenge to get high results.
2- Also, when the databases are equal (e.g., depressed and non depressed databases), the results often become higher
2- Give some significant points in the conclusion section in terms of this area.
3- Write the paper in very good English.
4- Moreover, check out these articles may help you in this area:
https://www.sciencedirect.com/science/article/pii/S153204641830090X
Thesis ''Automatic assessment of depression from speech: paralinguistic analysis, modelling and machine learning''
https://www.sciencedaily.com/releases/2019/07/190712133338.htm
https://ieeexplore.ieee.org/abstract/document/9077698
And all the best (:
You can publish works that had poor results. However, you must justify why it had such poor results. The underlying theory must be presented and well discussed. Having those kinds of research are useful as others that had successful results. Just make sure that your analysis is well validated and your proofs are strong enough to support that its not the fault of improper implementation. Good luck my friend. :)
Dear nik nur wahidah nik hashim ,
Of course you can publish it following recomendations of Francis Jesmar Perez Montalbo
Howerver, in machine learning context, the published works get better performance than others or present new aplications.
Thank you for all the responses. I am currently trying to validate the result with making sure the algorithm works well (testing it on another database that has been known to have good classification/prediction result). Also currently trying to gather another database in order to see if the result is still the same. Writing paper for negative outcome is more challenging than writing for positive output analysis. Will do my best ! Thanks again :)
Hello Dr. Nik!
No result is also a result. So a 'poor result' should not deter you from publication if you manage to convey its relevance. Consider the following -
1. Define the Research Question clearly. You could consider using the PICOS paradigm to do so.
2. Dwell on the 'Gap in research' that your 'Research Question' addresses, at the end of your Introduction. It has to grab the reader's interest.
3. Ensure that your methods are un-ambiguously described so that your study is replicable.
4. The strength of the study can be enhanced if the comparison group (normals) is larger than the study group (depressed).
5. Selecting specific search items in speech (identified from previous studies) and evaluating for frequency etc. might enhance the hit rate/ accuracy.
6. Increasing leangth of search strings/ chunks might also improve hit rate/ accuracy.
7. The Results should be presented in easy-to-comprehend tables/ figures.
8. The Discussion needs to bring out what is new/ different and also what conforms with previous research/ a thorough review of literature. The possible reasons for a 'poor result' need to be made clear.
9. Finally, a clear Conclusion which highlights the way forward, even if it is a closed door, will make it more acceptable.
10. Apart from the merit of the content, the readability/ flow/ interest quotient will be used by the editors/ reviewers to select the article.
Happy publishing!
Dr. K Chatterjee
Professor of Psychiatry
AF Medical College
Pune, India
- Badges
- Science topic