ICACT20210193 Question.1
Questioner: joomi@inje.ac.kr 2021-02-09 오후 2:19:00 |
ICACT20210193 Answer.1
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 2:19:00
Chrome  Click!! |
| Thank you for the slides. How machine learning techniques is beneficial compared to the traditional methods? |
Thank you for the question. In the traditional method detecting COPD takes long time and also need expertise to correctly identify different stages. However by using machine learning techniques, we could able to identify important features and develop algorithm that can automatically identifies different stages quickly in a precise manner. |
ICACT20210193 Question.2
Questioner: tanatthep001@gmail.com 2021-02-09 오후 3:49:36 |
ICACT20210193 Answer.2
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 3:49:36
Chrome Click!! |
| how many number of dateset in ESCP and ASCP? |
Thank you for the question. The number of patients in ESCP is 2300 and ASCP is 600. We have used oversampling technique for solving the imbalance problem. |
ICACT20210193 Question.3
Questioner: tanatthep001@gmail.com 2021-02-09 오후 3:53:09 |
ICACT20210193 Answer.3
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 3:53:09
Chrome Click!! |
| If we would like to detect the different stages of COPD by using more than two classes, you think can SVM do that and what technique should be applied if it has more than two classes? |
Thank you for the question.Yes SVM could able to detect the different stages of COPD if there are more two classes . Although SVM is mainly used for binary classification problem, it can be used for multiclass classification problem based on "One-vs-Rest" strategy which is Fit one binary classification model for each class vs. all other classes. |
ICACT20210193 Question.4
Questioner: franklinlu888@outlook.com 2021-02-09 오후 4:22:20 |
ICACT20210193 Answer.4
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 4:22:20
Chrome Click!! |
| In your opinion, will the pre-process to find the features of the COPD signals aid you to recognize the respiration sounds? |
Thank you so mcuh for the question. As per your question, we can detect the respiration sound using COPD signals. However we have included that part in our present study. This question is out of our scope for the present study. |
ICACT20210193 Question.5
Questioner: quandh@vnu.edu.vn 2021-02-09 오후 5:50:21 |
ICACT20210193 Answer.5
Answer by Auhor satyabrataaich@gmail.com 2021-02-09 오후 5:50:21
Chrome Click!! |
| Can you please explain more about your data set, since I wonder they are an image-based dataset or not?
In case of image-based data set, I wonder if you perform your algorithm on the whole image or just parts of its since COPDs could be very small and appear only on few slices.
As a consequence, you had better provide more details of feature extraction techniques that were used in your work. |
Thank you for the questions.No. Our dataset consists of clinical data. We are not considering any image data for detection of COPD for this proposed study. |
ICACT20210193 Question.6
Questioner: 17862328070@163.com 2021-02-10 오후 6:06:18 |
ICACT20210193 Answer.6
Answer by Auhor satyabrataaich@gmail.com 2021-02-10 오후 6:06:18
Chrome Click!! |
| I'm very interested in your research and have two questions to ask. The first question is that in your experimental comparison method, you have used traditional machine learning methods, but have you considered using deep learning methods, such as neural networks for disease classification? |
Thank you for the question. To be honest we have applied the traditional methods but not applied any deep learning method in this study. In future we will do it. Thank you for the suggestion |
ICACT20210193 Question.7
Questioner: 17862328070@163.com 2021-02-10 오후 6:08:09 |
ICACT20210193 Answer.7
Answer by Auhor satyabrataaich@gmail.com 2021-02-10 오후 6:08:09
Chrome Click!! |
| The second problem is that the F1score you used to evaluate the model is indeed a metric that can weigh the detection rate and accuracy of the model, but there is a default premise that the weight of the two types of errors committed by the model is the same, that is, we default to the model to predict the disease that is not this disease as this disease and the prediction of the disease that is not this disease has the same weight. However, in practice, this is often seen as a non-equivalence, have you considered this problem? |
Thank you for the question. This is a great question and to be honest we did not think about it.But we will consider this point and then work on it to solve these kind of problem. |
