Dear Hadeel,

To tackle your problem, your first need to characterize the domain of your study, your objective and your data. Medical Domains are extremely broad and challenging as well… I’ll leave some “food-for-thought”, which may help.

Frequent tasks in medical domains are as follows:

• Diagnosis: To detect some disease, usually a classification task (binary or multi-class, depending on the possible outcomes).
• Treatment: You may also try to “recommend” the best treatment option for each patient, if you have this information (classification).
• Recurrence: Patients affected with some disease may overcome it, but fall ill again. This is also an important field of study in medical domains. Time for recurrence for instance, or site of recurrence (classification).
• Survival Prediction: Survival Prediction is extremely important because it may influence the choice of important end-of-life decisions for doctors and patients (classification).
• Subgroup Discovery: Patients may be “divided” in somewhat homogenous groups (clusters). These clusters may be associated with different outcomes, different responses to treatment, and you may want to determine which makes the patients of each cluster similar, discovering rules that define a “type” (subgroup) of patient. Some medical guidelines are often based on staging systems that may not account for the entire patient information. With clustering, you can provide a different or perhaps more complete perspective for doctors.
• Knowledge Discovery (feature selection or rule induction): You may want to find correlations between variables in data, or determine which features are truly impacting the outcome.

You may find some examples for the case of breast cancer:

Article Predicting Breast Cancer Recurrence Using Machine Learning T...

Besides the most common tasks, I would like to add that medical datasets are typically challenging domains because of the importance of medical predictions and also due to the inherent complexity of data.

• Explanability and Interpretability: This is a hot-topic in bioinformatics since, to be adopted by the medical community, machine learning algorithms need to be both reliable (high performance results, high accuracy), and interpretable (one cannot jeopardize the other).

  Conference Paper Interpreting Deep Learning Models for Ordinal Problems

• Complex Data:

Hope this helps!

Have a look at :

Article Machine learning in medicine: a practical introduction

https://www.thieme-connect.com/products/ejournals/pdf/10.15265/IY-2015-014.pdf

https://medium.com/sciforce/top-ai-algorithms-for-healthcare-aa5007ffa330

Article Machine Learning in Medicine

I recommend

Chapter Machine Learning in Medical Applications

Article Machine learning for medical diagnosis: History, state of th...

Article A new feature selection method on classification of medical ...

Article Classifier ensemble construction with rotation forest to imp...

Article Machine Learning in Medicine

Article A hybrid intelligent system for medical data classification

Article Machine Learning in Medical Imaging

Conference Paper Batch Mode Active Learning and Its Application to Medical Im...

https://onlinelibrary.wiley.com/doi/abs/10.1002/1097-0142(20010415)91:8%2B%3C1636::AID-CNCR1176%3E3.0.CO;2-D

http://www.jhi-sbis.saude.ws/ojs-jhi/index.php/jhi-sbis/article/viewFile/226/166

https://www.sciencedirect.com/science/article/pii/S0098135417302570

Medical imaging is an emerging area where machine learning is used to replace doctors for diagnosis of diseases. Medical images like normal photos, X-rays, cardiograms, ultra-sonograms etc. are fed to classification algorithms to differentiate between normal and diseased cases, usually in a supervised way.

Hadeel S. Obaid based on your objective and reseach goal, you can perform classification, clustering on respective dataset. You can have a look on below links for better understanding:

Article Medical Dataset Classification: A Machine Learning Paradigm ...

https://ieeexplore.ieee.org/abstract/document/7977277

Article The distance function effect on k-nearest neighbor classific...

https://www.researchtrend.net/ijet/pdf/Analysing%20the%20Performance%20of%20Classification%20Algorithms%20on%20Diseases%20Datasets%20LAXMI.pdf

https://asmedigitalcollection.asme.org/medicaldiagnostics/article/3/1/011004/975459/Applying-Machine-Learning-Methods-Toward

Hope this research work will gave you idea.

Classification is particularly useful in predicting binary outcomes in medical data sets. You can employ a number of different ensemble methods, neural networks, and other algorithms to provide clinically meaningful information. Please see some of these helpful links below:

Article Machine Learning in Medicine

Article Machine-learning to stratify diabetic patients using novel c...

https://pubmed.ncbi.nlm.nih.gov/31983042/?from_term=prem+machine+learning&from_sort=date&from_pos=1

Edx also has free introductory courses on machine learning theory and how to use it in programs

https://www.edx.org/course/subject/data-analysis-statistics

Classification is mainly used in supervised learning methods where predefined labels are assigned and the trained model is to predict binary outcomes (such as true or false), whereas clustering is used in unsupervised learning where similar instances are grouped, based on their features or properties (there are no predefined class labels).

With the help of machine learning we can use the data for disease Identification also real time data analysis will help in motor restoration, locomotion, and communication.

In biosignals and classification there, beside ML-methods and deep learning methods, it is occurring very useful to apply entropy measures as the way of feature extraction from the native data.

In such a case, I do not recommend any kind of signal preprocessing as it can 'erase' an important part of the original data that is useful within entropy evaluation.

It is an exciting area to be as it will make great progress in the coming years. Definitely not easy, mathematically demanding, but absolutely worth it.

Good luck in your search :-)

I join this discussion and ask: how do I determine what the research problem is in the two sets of medical records I have?

Datasets are attached.

What is a research problem?

What should be investigated on these medical datasets?

The datasets were taken from machine learning repository as Hepatitis C Virus (HCV) for Egyptian patients Data Set.

You can do exploratory analysis, clustering the age group, regression too

Ankit Nayan

Thank you very much!

- What would be exploratory analysis and how is it performed?

- How would age grouping be done?

- How would a regression be derived and what would be obtained from it?

Best regards,

Željko Vujović

Welcome

1)Please check this article for exploratory analysis :-

https://www.google.com/amp/s/www.geeksforgeeks.org/exploratory-data-analysis-in-python/amp/

There are more such articles on website like a)towards Data science

b) being datum

2)By help of clustering algorithm you can group the particular age of people having diseases in common.

Ankit Nayan

I need to commpare the reliability measures of WEKA machine learning methods: accuracy, precision, sensivity (Recall) and specifity, using the HCV-Discretization-Criteria.csv dataset I posted above in the comment.

What do you recommend for me:

how do I compare these reliability measures?

Kind regards,

Zeljko Vujovic

hi

to use the data set with classification and matching process

Depending on

1- Your dataset type such as images,..., CSV files.

2- If the data with it's labels or not (supervised or unsupervised learning).

3- The objective you want from the dataset.

It all depends on the data you have and the objective of your project.

1) For example, if the dataset has been classified already (labeled), u may want to apply classification algorithms

2)If the dataset has no target variable, try to cluster it.

3) If the dataset has a continuous target variable, go for Regression.

If the problem at hand was described in detail, a more precise solution would have been possible.

I think that today when we have intelligent drug, we have to work with fuzzy entropy and fuzzy neural networks.

You can create an adaptive system of diagnostics and recommendation system. If you have the data of the clinical picture, anamnesis and tests, as well as verified diagnoses (reliable), then this is not a problem. I can show You how it's done

Dear Eugene Veniaminovich Lutsenko , thank you for your reply.

Actually I got a medical data set. The quality of data is very bad. At the moment, I am cleaning the data. I would like to analyse the data using clustering or classification in such a way it can help in medical field.

The automated system itself must clear the data. If it's white noise, it doesn't matter at all. If the error is systematic, it will not affect the conclusions. Look at my new work

If you send me a small fragment of your data, I'll see what I can do. It is necessary that you formulate the task. If I look at the data, I can tell what tasks can be solved based on it. Maybe I'll make a pilot model that you can expand with additional data. I strongly do not recommend improving the data. You can distort them. It is desirable to process the data "as is".

Dear Hadeel,

To tackle your problem, your first need to characterize the domain of your study, your objective and your data. Medical Domains are extremely broad and challenging as well… I’ll leave some “food-for-thought”, which may help.

Frequent tasks in medical domains are as follows:

• Diagnosis: To detect some disease, usually a classification task (binary or multi-class, depending on the possible outcomes).
• Treatment: You may also try to “recommend” the best treatment option for each patient, if you have this information (classification).
• Recurrence: Patients affected with some disease may overcome it, but fall ill again. This is also an important field of study in medical domains. Time for recurrence for instance, or site of recurrence (classification).
• Survival Prediction: Survival Prediction is extremely important because it may influence the choice of important end-of-life decisions for doctors and patients (classification).
• Subgroup Discovery: Patients may be “divided” in somewhat homogenous groups (clusters). These clusters may be associated with different outcomes, different responses to treatment, and you may want to determine which makes the patients of each cluster similar, discovering rules that define a “type” (subgroup) of patient. Some medical guidelines are often based on staging systems that may not account for the entire patient information. With clustering, you can provide a different or perhaps more complete perspective for doctors.
• Knowledge Discovery (feature selection or rule induction): You may want to find correlations between variables in data, or determine which features are truly impacting the outcome.

You may find some examples for the case of breast cancer:

Article Predicting Breast Cancer Recurrence Using Machine Learning T...

Besides the most common tasks, I would like to add that medical datasets are typically challenging domains because of the importance of medical predictions and also due to the inherent complexity of data.

• Explanability and Interpretability: This is a hot-topic in bioinformatics since, to be adopted by the medical community, machine learning algorithms need to be both reliable (high performance results, high accuracy), and interpretable (one cannot jeopardize the other).

  Conference Paper Interpreting Deep Learning Models for Ordinal Problems

• Complex Data:

Hope this helps!