Issues Regarding Dataset Preprocessing

[Rumi07]

Dear Author,

Thank you very much for sharing your excellent work. I truly appreciate your contribution and am very interested in exploring it further.

While going through the dataset preprocessing steps, I encountered some difficulties with a few of the datasets. Apart from the four datasets in CSV format that were already shared (arcene, isolet, micromass, and parkinson), I am having trouble preparing the CSV files for the others. I would greatly appreciate your guidance on the following:

• For the Lung and TOX-171 datasets, I noticed that the shared files are in .mat format. Should I convert these .mat files into CSV format for use in the code?

• For the BCTIL dataset, there are several files including cluster.txt, file_suplemental_info.txt, metadata.tsv, normalised_data_unimputed.tsv, trm_counts.tsv, trm_saver.tsv, and trm.h5seurat. Could you kindly let me know which of these should be used for preprocessing, or if any of them should be converted to CSV?

• Regarding the P53 dataset, after downloading from the provided link, I found two folders (p53_new_2012 and p53_old_2010), each containing files like K1-K9.def, K9.data, p53.names, and K8.instance.tags. I would be grateful if you could share which files should be used and how I might convert them to CSV format for use with your code.

• Lastly, for the HAR dataset, could you please let me know which specific file should be used for preprocessing?

Thank you again for your time and support. Any help would be greatly appreciated. I look forward to your guidance.

[rui-yan]

Hi, thank you for your interest in our work.

To address your specific questions:

• Yes, the datasets should be converted to CSV format.
• The features are in normalised_data_unimputed.tsv, and the labels are in metadata.tsv.
• The p53_new_2012 K9.data file is used.
• For HAR, we subsampled the data by randomly selecting 10% of the samples, as specified in the dataset section, followed by standard data preprocessing, as implemented in the data_preprocess function in this file.

[Rumi07]

@rui-yan, thank you so much for your response; it was really helpful.

• Would you please share what preprocessing you used for the QSAR and WDBC datasets?
• In the WDBC dataset, there is a file named wdbc.data, for which preprocessing should I follow for this one to convert the CSV file for the code.
• In the QSAR dataset, there is a file named biodeg.csv. Is it the data_raw.csv file that you directly use in your code, or did you use some preprocessing for this dataset?

Thank you again for your time and support.

[rui-yan]

For all the datasets, we converted original data into a data_raw.csv format, where each column corresponds to a raw feature (with feature names as headers), followed by a label column named Class. Some original files may include an ID column—these should not be treated as features.

The original data for WDBC is wdbc.data, and for QSAR is biodeg.csv.