A New Stock Price Forecasting Method Using Active Deep Learning Approach

Round 1

Reviewer 1 Report

The authors describe how to estimate adjusted -close price of specific corporations. The paper reports the results of the accuracy with lower loss by creating six feature sets and using six deep learning models. This paper showed adding extra features improved the performance and LSTM model performed better compared to others.

Pros:
1) The paper is well motivated. It attacks a very relevant problem of improving the price of stock-price prediction.

Cons:
1) There is not enough explanation why LSTM gave better result and for some data sets 4 features gives higher accuracy than the six features.

Comments:
- This paper is well written, The authors explain very well the ML methods that they used, but they did not provide some relevant information on how the experiments are performed.
- I would like to understand: (1) Which are the most efficient ML methods in terms of execution time (training + inference)? (2) Which are the ML methods that require fewer training samples to be trained but they achieve the same accuracy as others?
- The authors should also discuss their limitations, and how the result can be improved in the future. 

Overall, I like the paper. I suggest a minor revision.

Author Response

Point: This paper is well written, The authors explain very well the ML methods that they used, but they did not provide some relevant information on how the experiments are performed.

Response: Thank you for your comment. We added the main requested information as follows.

 The CNN model showed the best efficiency in terms of execution time. GRU and CNN were the best models for giving good results with fewer examples.

Point: I would like to understand: (1) Which are the most efficient ML methods in terms of execution time (training + inference)? (2) Which are the ML methods that require fewer training samples to be trained but they achieve the same accuracy as others?

Response: Thank  you for your comment.  The CNN model showed the best efficiency in terms of execution time.

Point: The authors should also discuss their limitations, and how the result can be improved in the future. 

Response: Thank  you for your comment.  The research limitations include: depending only on a basic deep learning model, the research did not investigate using transformer-based approaches or transfer learning, and finally, the research area of time series analysis does not have a big pre-trained model like BERT in NLP, and DALL-E2 in the computer vision domain. Thus, this area might be covered better in the future work.

Overall, I like the paper. I suggest a minor revision.

Thank you very much and your effort are highly appreciated.

Reviewer 2 Report

Please find the comments on attached file.

Comments for author File: Comments.pdf

Author Response

Point: . The title mentioned “Active deep learning” but throughout the paper, the term “active” is not used. Why?
Response: Thank you for your comment. We mentioned active as we define the best deep learning method as an active method from the comparative methods in this paper.
Point: The abstract is not written systematically. It needs to present the problem, gaps, contribution, and results. Authors repeated the same phrase “this study also investigates…” in more than two places in the abstract.
Response: Thank you for your comment. We revised the abstract section to make it clearer.
Point: . The sentence “A comparison between deep LSTM, MLP, ELSTM deep learning models is introduced in [26] and the SVM machine learning model and linear regression “ can be well written as “The performance of various deep learning models such as deep LSTM, MLP, ELSTM models in [26], LSTM, GRU in [1] and the SVR and NN in [2] were compared for stock price forecasting”. ([1]DOI: 10.3390/math8091441) ([2] DOI: 10.1109/ICAECC.2018.8479456)  
Response: Thank you for your comment. We revised it as suggested and added the given references.
Point: The sentence “Deep learning models proved to give excellent results in many areas, and they showed a potential of being used in stock market prediction …. “ can be written as Deep learning models proved to give excellent results in many areas [3-4], and they showed a potential of being used in stock market prediction…..”. ([3]DOI10.1371/journal.pone.0264586) ([4] DOI: 10.1038/s41598-021-03287-8)
Response: Thank you for your comment. We revised it as suggested and added the given references.
Point: The sentence “It improves vanilla Recurrent Neural Network (RNN) [11]” needs further clarification or reference.
Response: Thank you for your comment. Done
Point: The details about implementation tools such as colab, python, etc. should be written in a separate subsection rather than as methodology.
Response: Thank you for your comment. In the results section, we added “This research utilized python programming language and was performed using Google-Colab, many libraries have been used like: Pandas, Numpy, Matplotlib, Sklearn, Keras.”
Point: More evaluation metrics such as coefficient of determination (R2 ) need to be compared.
Response: Thank you for your comment. According to the limetted time and current situation, we can do this new measure in future research.
Point: A thorough analysis and discussion of results need to be organized.
Response: Thank you for your comment. We tried our best to rearrange the results section and make it clearer for the readers.
Point: The overall flow diagram of the proposed work needs to be presented.
Response: Thank you for your comment. We added a new figure to show the overall flow diagram of the proposed work
Point: Writing sentences like “As shown in Figures 6 and 7 above” is not standard as the figure might move below or above while typesetting.
Response: Thank you for your comment. Revised
Point: .English needs to be improved, especially the use of tense.
Response: Thank you for your comment. The language of the whole paper has been checked by a native speaker researcher
Point: Analysis and discussion are not enough. There are many bar diagrams but little explanation.
Response: Thank you for your comment. We added more explanation as requested.

Round 2

Reviewer 2 Report

The authors attempted to address most of the issues raised in the previous round but not all. However, there are minor issues still presented in the manuscript. I have the following suggestions for the authors.

 Comments

1.       The main contributions of the proposed work need to be explicitly mentioned. Better to present it in the form of a list of contributions.

2.       Line 105-106 has no proper citation style. It should be in the format of the author (date) or author et.al. [number]

3.       Subscripts and superscripts used in various places need to double check. There are a few errors.

4.       The description of models such as CNN, LSTM and so .on. need more logical organization. It seems the content is scattered too much. For example, in lines 297-298, this research introduces….., why author introduce their model in the middle of the model description? Better authors introduce the model in a sequence, first, MLP, CNN, LSTM, BI-LSTM, and others.

5.       Typesetting needs to be examined properly. For instance, a bullet in the GRU model, or LSTM model … needs to remove.

6.       Line 402-404, many libraries have been used…., how many? Please be specific rather than general.

7.       Please pay serious attention to carefully correcting the typos and English errors.

Author Response

Point: The main contributions of the proposed work need to be explicitly mentioned. Better to present it in the form of a list of contributions.

Response: Thank you for your comment. We present the main contributions of this paper as points. The main contributions of this paper are presented as follows.

• Study the effects of the additional features (i.e., High, Low, Volume, Open, HiLo, OpSe).
• Detect the effect of the size of the datasets on the prediction accuracy.
• Detect the difference between the deep learning models (i.e., MLP, GRU, LSTM, Bi-LSTM, CNN, and CNN-LSTM).

Point: Line 105-106 has no proper citation style. It should be in the format of the author (date) or author et.al. [number]

Response: Thank you for your comment. We added a reference as requested.

Point: Subscripts and superscripts used in various places need to double check. There are a few errors.

Response: Thank you for your comment. We checked the mathematical notation of this paper.

Point: The description of models such as CNN, LSTM and so .on. need more logical organization. It seems the content is scattered too much. For example, in lines 297-298, this research introduces….., why author introduce their model in the middle of the model description? Better authors introduce the model in a sequence, first, MLP, CNN, LSTM, BI-LSTM, and others.

Response: Thank you for your comment. We changed the order as requested.

Point: Line 402-404, many libraries have been used…., how many? Please be specific rather than general.

Response:

Point: Please pay serious attention to carefully correcting the typos and English errors.

Response: We checked the whole paper and make sure the language is clear and correct.