My research on classification algorithms lead to me lot of conclusions. From my conclusions I would say that SVM is the better classification algorithm.

SVM is an interesting classification algorithm. However, based on the model being used ANN can provide tough competition to SVM. A comparison of both is subject to multiple argument as to what should the benchmark be. In some benchmark it has been observed that SVM outclass ANN while in others it is the reverse. So, I personally believe that the data to be classified is an important parameter to be considered when choosing a classification algorithm.

relevance vector machines

they are definitely better classifier as they use probabilistic approach

also belief propagation algos have not been fully exploited 

Of course it is not. There are several parameters that affect the classifier's performance such as distribution of training data  and the parameters to run the algorithm etc. Since these reasons above, we can not say it is the best. It may be better than the others however it is really based upon the application implemented.   

Dear Rangarajan,

Its up to your problem especially it depends on the form of your data problem. For example, for data that are like as spars vectors and matrices, classifiers based on SVD and HOSVD are better than the other. Generally, stating like what you said, is not always true. SVMs are good but as I said it depends on your problems.

best regards,

Hamid 

I do agree with all the above answers which confirm that SVM is an interesting classification methods, yet its hard to generalize its superiority for all classification tasks. based on my experience, the answer o the question "which classification method is the best performer?"  is task-dependent. One more thing you need to consider for answering this question, that is the setting of the parameters for the classification method plays a vital role in determining its performance.        

Sirs thank you so much for the quick and valuable responses. I will try for my best by following your guidelines.

In any case SVM  is one of the best classifiers. During my experimentation with financial and economic datasets, most of the times SVM outperforms the rest (but not always). So, going with the flow on this debate I'll say that there can't be a simple yes or no answer to your question.

In machine learning, the classifiers performance is really based upon the distribution of training&testing dataset and the parameters required for running the kernel functions.  Just see the article below :

https://openaccess.leidenuniv.nl/bitstream/handle/1887/13575/Chapter%207.pdf?sequence=12

https://openaccess.leidenuniv.nl/bitstream/handle/1887/13575/Chapter%207.pdf?sequence=12

there is no free lunch and no algorithm perfoms always best.  The most complete up-to-date comparison can be found in the link below. In that work SVM with RBF kernel appears in second place in overall performance after random forest using CARET (R)

http://jmlr.org/papers/volume15/delgado14a/delgado14a.pdf

Whenever you know distributional nature of features you are interested in, it is a good option. It is because generally you have a set of families of kernels. However, even if you know distributional nature of data, it is not garanted the available family of kernels that fit the spectrum of data. In that cases, it is ensured best beahviour of any adaptive algorithm e.g. Neural Networks; even without know the spectrum of data. I hope it could be helpful.

Although, SVM has shown a superior performance over other classification methods in many applications,  it is hard to generalize this to all classification problems. In fact , to me the question "which classification method is the best ?" can not be answered in isolation of the problem (data) specifications and the settings of its parameters. What it can be generalized is that the performance of any classification method is problem-dependent and the setting of its parameters play a vital role in this regard. 

Thank you sirs for your valuable responses. I have started to work with support vector machines

There is no always winning algorithms in machine learning. See "no free lunch theorem" for more details.

http://www2.ece.ohio-state.edu/~aleix/Algorithm-Independent_Learning.pdf