Usually, machine learning and data mining problems suffer from the so called 'No free lunch theorem' (http://en.wikipedia.org/wiki/No_free_lunch_theorem). Basically, it states that there is not a single machine learning algorithm that outperforms the others for every single domain. Putting it simple, you will always find domains where some algorithms excel, while they will perform poorly in others. It all depends on the nature of your data and the problem being solved at hand. Hence, you will need to identify the characteristics of your data and apply the appropriate algorithms that fit such characteristics.

With regards to the software tools that you may use to solve data mining problems, there are many software packages. Some of the most popular are Weka (http://www.cs.waikato.ac.nz/ml/weka/), Mahout (http://mahout.apache.org/), Scikit-learn for Python (http://scikit-learn.org/), R's  set of machine learning packages, etc. You can find a funny visualization of some of the most important machine learning libraries at http://www.mln.io/resources/periodic-table/.

I hope it helps you!

Dear Linden

What is difference between health care data and other data?if there is no difference,answer will be the same for other data.I think some differences e.g security issues does not change the answer.Two article maybe useful for our purpose:

1-Uniqueness of medical data mining

KJ Cios, GW Moore - Artificial intelligence in medicine, 2002 - Elsevier

2-Selected techniques for data mining in medicine

N Lavrač - Artificial intelligence in medicine, 1999 - Elsevier

Abstract

Widespread use of medical information systems and explosive growth of medical databases require traditional manual data analysis to be coupled with methods for efficient computer-assisted analysis. This paper presents selected data mining techniques that can be applied in medicine, and in particular some machine learning techniques including the mechanisms that make them better suited for the analysis of medical databases (derivation of symbolic rules, use of background knowledge, sensitivity and specificity of induced descriptions). The importance of the interpretability of results of data analysis is discussed and illustrated on selected medical applications.

http://www.sciencedirect.com/science/article/pii/S0933365798000621

Thank you for your response, Hamid.

In general, I would argue that there is no difference between medical data and data from other industries, as it relates to the data mining process. Security is not relevant here, as these data are already accessible to the analyst and have passed the appropriate "security clearance".

I am also interested in which software is typically used (or recommended) to implement the various (and recommended) data mining techniques.

All suggestions are welcome.

Ariel

Usually, machine learning and data mining problems suffer from the so called 'No free lunch theorem' (http://en.wikipedia.org/wiki/No_free_lunch_theorem). Basically, it states that there is not a single machine learning algorithm that outperforms the others for every single domain. Putting it simple, you will always find domains where some algorithms excel, while they will perform poorly in others. It all depends on the nature of your data and the problem being solved at hand. Hence, you will need to identify the characteristics of your data and apply the appropriate algorithms that fit such characteristics.

With regards to the software tools that you may use to solve data mining problems, there are many software packages. Some of the most popular are Weka (http://www.cs.waikato.ac.nz/ml/weka/), Mahout (http://mahout.apache.org/), Scikit-learn for Python (http://scikit-learn.org/), R's  set of machine learning packages, etc. You can find a funny visualization of some of the most important machine learning libraries at http://www.mln.io/resources/periodic-table/.

I hope it helps you!

Thank you, Victor. This is helpful.

So as a follow up, how does a researcher decide which machine learning algorithm to apply? Is there some measure of "fit" as there is in statistics with AIC, etc.? How would someone determine which algorithm to start with, and where to go from there?

Dear Linden,

I totally agree with Hamid, 'Data is data' and it all depends on what you want to do. You have a variety of statistical tools to choose from basically depending on your objectives/research questions and/or hypothesis set.

For a slightly different response, Ariel, the best data mining tool is the mind of a talented statistician.  Whatever package that person uses to address your questions is their best resource, but the idea that asking a machine to do things automatically instead of a statistician to do things thoughtfully with sensitivity to the specific issues in your sample and assessment is a no-brainer choice to me. 

Depending on your problem, one may use some metrics or others. For instance, in a classification problem, one usually uses the prediction error rate as a measure of the quality of the algorithm. In contrast, the root mean squared error can be used to assess the quality of models in regression problems.

In general, testing a machine learning algorithm against a dataset involves partitioning your data into a training set and a test set. The model is trained with your training set, whereas the model accuracy is tested with the test set. One of the common approaches in the field is what is known as cross validation (http://en.wikipedia.org/wiki/Cross-validation), which consists of dividing your data into n disjoint partitions. Then, n-1 partitions are used as training and 1 partition is used as test. The process is repeated until each partition has been part of the test. This way, one can avoid unlucky partitioning and it is possible to get a more accurate measure on the quality of the model in a real setting.

Another common approach in the field is employing an Occam's razor strategy. First, the simplest algorithms and evaluated. If the accuracy of these algorithms is unacceptable, one can start to test more sophisticated algorithms. The rationale behind this idea is that, in equal conditions of performance, models with a lesser number of parameters are preferred over those with a greater number due to the latter being more prone to overfitting.

If you are a newcomer to datamining and machine learning, may I recommend taking Stanford's MOOC on Machine Learning (https://www.coursera.org/course/ml)? Now it can be done at your own pace! 

So as a follow up, how does a researcher decide which machine learning algorithm to apply? Is there some measure of "fit" as there is in statistics with AIC, etc.? How would someone determine which algorithm to start with, and where to go from there?

Try see SAS and Mathematica

Hi Ariel,

In the last decades we have witnessed a technological evolution that allowed us "look" at patients in unprecedented way. Some of this technology was miniaturised and made wearable, leading to a real-time quantified self. Today, the future seems to be anti-wearable, open-source, based on Do it Yourself platforms and Powered by the Community. The multi-dimensional data that each of us is able to generate daily will not only challenge the best and most efficient data storage and mining tools, but also how this data will help support clinical decision making after being distilled. Technology will be certainly pushed further but we also need to push health-related courses' curriculum to integrate curricular units where clinicians can develop knowledge and competencies on how to use/interpret this information/systems.

There are a wide variety of free/paid tools that can be used in this context; which to choose depends on what is the question you want to answer, your stats/machine learning/programming/... degree of knowledge, how you want to output the data, and many other things.

Assuming that someone would like to start mining health-related (big) data, I would suggest a more user-friendly visual environment with no(t so much) programming requirements like:

- Orange: open source data mining package;

http://orange.biolab.si

- Weka: an open source stats and data mining library;

http://www.cs.waikato.ac.nz/ml/weka/

- RapidMiner: an open source statistical and data mining package;

https://rapidminer.com

After some data mining experience and with some more programming skills:

- Scikit Leran

http://scikit-learn.org/stable/

- R

http://www.rdatamining.com

On a level II stage I would look into some big data analytics platforms and highly effective operational databases:

http://www.informationweek.com/big-data/big-data-analytics/16-top-big-data-analytics-platforms/d/d-id/1113609

http://voltdb.com/?n=a&utm_expid=66808973-1.O562mQS1TnKG3RbsGkKu5Q.1

Some good resources to start from (from my experience).

Python oriented:

- Programming Collective Intelligence http://shop.oreilly.com/product/9780596529321.do

Very popular among beginners with real-life examples (data)

Many Python examples

- Machine Learning an Algorithmic Perspective

http://www.amazon.com/Machine-Learning-Algorithmic-Perspective-Recognition/dp/1420067184

Mainly focused on algorithms

Not so much math and many Python examples

- Machine Learning in Action

http://www.amazon.com/Machine-Learning-Action-Peter-Harrington/dp/1617290181

Similar to the latter; algorithm hands-on

Many Python examples

Weka (platform) oriented:

- Data Mining: Practical Machine Learning Tools and Techniques

http://www.cs.waikato.ac.nz/ml/weka/book.html

Very good introduction on ML

It is a companion book for the Weka platform

I would also look into:

- Tom Mitchell's book - Machine Learning (1997) - and online material

http://www.cs.cmu.edu/~tom/

- Christopher Bishop's - Pattern Recognition and Machine Learning (2007)

http://www.springer.com/gp/book/9780387310732

- Jason Brownlee's - Machine Learning Mastery web page.

http://machinelearningmastery.com

I hope it helps!

Hi Ariel:

Data mining is an area I am interested in but have not studied/researched it, so I looked up what is available on ResearchGate:

https://www.researchgate.net/publicliterature.PublicLiterature.search.html?type=keyword&search-keyword=data+mining+tools+for+health+care+data&search-abstract=&search=Search

Best regards,

Debra