What data you are trying to mine.

It depends on the nature, size and quality of data.

It depends on your data, if your data is medical image, fuzzy logic and Support Vector machine are suitable techniques. Especially, if there are existing noise in your data, Type-2 fuzzy classification is more better.

If your data is time series then you should take a look at Eamonn Keogh's website. In most of the cases, it is not important what classifier you should use. It is more important how you understand the data and extract good and rich feature set from the data.

I think it is not a question of whether data is from medical field or not, but the nature of the data. You may like to elaborate the kind of data you are dealing with -- images, numeric values, qualitative values, etc -- as well as their uncertainty/reliability aspects. Also the nature of classification you are interested in will be useful -- binary vs multi-class, overlapping vs mutually exclusive, etc. Classification is a very very researched domain -- plenty of techniques and experiences can be found in literature.

HI

it depends on the data type that you are working with.

There are many in healthcare and medical field. You can try "science direct" website".

Research on clasifying the clinical data sets is going on for more than a decade. Infact there are standard data sets that are available for testing new algorithms to classify the clinical data.

Hello,

You can use the map-resuce paradigm with the Apache Hadoop. Just take a look at the linux Cloudera distribution and follow two links beneath:

http://www.teradata.com/white-papers/Hadoop-and-the-Data-Warehouse-When-to-Use-Which/

http://publications.hevs.ch/index.php/attachments/single/433

It strongly depends on your type of data. You can not find a general technique for everything. You have to try different algorithms and chose the best one based on your application.

There are some challenges to deal with in the medical field with regards to discrimination of symptoms diseases, etc. These are:

1)Is the data imbalanced? For example, how many people in your data set have cancer vs those who do not? One possible statement that could be made is that diseases are outliers of the general population. (for example, outbreaks of Botulism in the US vs the general population records there)

2) Has there been a causal relationship established or is it just correlation between variables on which your judgement relies about the performance of your baseline.

3) The characteristics of the data that have been gathered, are they enough to discriminate the disease?

4) Is the available data tagged or not. This will dictate whether supervised or unsupervised data is used.

These will determine the type of technique utilized as well as the validation level required to establish confidence in the procedure carried out.

As most people have already pointed out, in order to make a proper choice of a classification method / model, one must first understand the data well and different approaches work well for different types of features and class distributions.

Of course, there is no easy way to determine the best approach in advance, so - all theoretical considerations aside - one should always test several methods and evaluate their performance on some benchmark data sample.

In the medical field, class imbalance is often an issue, so it is important not only to look at the classification methods themselves, but also sampling methods (under-sampling, over-sampling, etc.). Additionally, one needs to be careful when evaluating and interpreting the results, as accuracy is no longer suitable, F1 score and the Matthews correlation coefficient need to be used instead.

Fuzzy methods would work well when there is noise in the data and there have been some new approaches to fuzzy classification recently. Also, SVM would be a good out-of-the box method, though I would still prefer decision trees and graphical models, for the sheer benefit of interpretability, as it gives some insight into the correlations (if not causal relationships) between the attributes.

SVM classification has highest priority following with neural network and genetic algorithm

Sometimes, improved data quality is itself the goal of the analysis, usually to improve processes in a production database and the designing of medical decision support. real life medical data often not is good quality ! I would suggest you to look in to the data pre processing prior to select any classification model. Data mining is a trial and error process and there is nothing called best !!! Its always domain and data specific. some methods maybe good for some domain and other may not. these articles may help you to understand the problem. 

Conference Paper Cluster Based Under-Sampling for Unbalanced Cardiovascular Data

Article Machine Learning Based Missing Value Imputation Method for C...