I m working on my dessertation using AI methodology in MATLAB.I have a text dataset so need some suggestions.
The performance of an algorithm (Deep learning or any other) has absolutely nothing to do with the format of the data. If you use a toolbox, you just have to ascertain that the toolbox can read csv of text data. Most toolboxes do, of course, but if your toolbox does not, it is your job to put your data in the appropriate data format.
The performance of an algorithm (Deep learning or any other) has absolutely nothing to do with the format of the data. If you use a toolbox, you just have to ascertain that the toolbox can read csv of text data. Most toolboxes do, of course, but if your toolbox does not, it is your job to put your data in the appropriate data format.
Reena Popat Deep learning works with many format of data.
What it is not good at is listed below:
- It doesn't work so well with small data. To achieve high performance, deep networks require extremely large datasets.
- Deep Learning in practice is hard and expensive. Deep learning is still a very cutting edge technique.
- Deep networks are not easily interpreted.
- Data imbalance is another problem that Deep learning faces.
Without analyzing the data it will be difficult to recommend what tool/software/algorithm can be best for it.
Start analyzing your data deeper and deeper, rest will come automatically.
I would say data and business problem is more important than tool/software/algorithm etc.
It is almost impossible to suggest a ML model based on the name of the dataset. The performance of the model depends on the data you are using. CSV or txt are data formats. Try to explain what's on the dataset.
DNN might or might not work for your data. It's impossible to suggest anything without knowing the data.
In DNN you need not to select features but in traditional ML you so have to select the features yourself. However, you need to have sufficient date and proper model to build an effective DNN.
I have EEG dataset in Text format. but very Small dataset .@Sakib Mostafa
For EEG I would suggest you to use LSTM, but the size of the data will be a problem. Better to use the traditional ML algorithms.
Reena Popat
By contrast to some answers given above, I would definitely not venture to advise you to use this or that algorithm without having the slightest idea about (i) the available data (except that it is in .txt format, which is completely irrelevant to the choice of the algorithm, as mentioned in my previous answer), and (ii) what you intend to do with it. So let me ask two sets of questions.
1) How small is your EEG dataset? I mean: how many subjects were recorded, how long are the recordings, at which sampling frequency were the recordings performed, how many channels were recorded ?
2) What task do you intend to perform by machine learning? Computer-aided diagnosis? EEG-based brain-computer interface? Artefact rejection? Should the task be performed in real time?
A caveat: there is a huge literature on machine learning for EEG data, and a huge amount of hype on deep learning for EEG; if your data set is small, you should just forget about deep learning anyway.
1) In general SVM is better for text data: https://medium.com/@bedigunjit/simple-guide-to-text-classification-nlp-using-svm-and-naive-bayes-with-python-421db3a72d34
2) Deep learning procedure is much better for the text data: http://openaccess.thecvf.com/content_iccv_2015/papers/Tian_Text_Flow_A_ICCV_2015_paper.pdf
Muhammad Ali
Reena Popat stated clearly above that the data are EEG data. EEG (electroencephalography) data are electrical signals recorded from electrodes placed on the scalp of subjects; they reflect the electrical activities of groups of neurons. Therefore the data is numeric, not textual. It is perfectly legitimate to provide numeric data in txt format (most of the numeric data available through the Internet on the Covid epidemic, for instance, are provided in text format). It is not clear what problem Reena Popat wants to address, but is is absolutely clear that it is not one of text classification, natural language processing, or the like.
For EEG I would suggest to architect your model with the 1-D-CNN because it will give you the best ACC and is suitable for small EEG datasets as well.
Naive Bayes and SVM classifiers can work well with textual data because they can handle sparse data case.
Text data can be processed using intelligent technologies. But we need software systems that provide this: http://lc.kubagro.ru/aidos/Works_on_ASK-analysis_of_texts.htm
Both algorithms are good for CSV data, but it depends on your data if you have enough data so deep learning is better than simple machine learning. otherwise, simple machine learning like SVM or ensemble SVM is good for you.
Hello all I successfully tried few algorithms and I have got more accuracy with SVM and ANN with my EEG dataset. Thanks for suggestions.
- Badges
- Science topic