Dear Gajendra,

SVM and k-means are very different. SVM is supervised (supervised classification) and k-means is unsupervised (clustering). so it depend on the goal of your application.

for supervised classification, SVM is the best algorithm and you need to precise je most efficient kernel (linear, RBF, etc...). for clustering, k-means is one of the simplest way to aggregate samples. For clustering, it already exist another approach such as Fuzzy methods. in the case of k-means two parameters needs to b taking account. the number of cluster a priori (classes) and the metric such as euclidian, mahalanobis, etc...

so you need to precise your goal!

best regards

Ludovic

There is no the best algorithm for all.  The type and selection of Attribute will influence in the results.

Classification and clustering are different tasks in data mining.

If you want to classify unknown samples into groups based on a target variable (classification), you may use SVM or other classification techniques such as artificial neural network, decision tree, k-NN, naive Bayes, or other ensemble techniques.

If you just want to group instances into clusters based on all variable without any target variable (clustering), you may use k-means which is the most popular algorithm for partitional clustering or you may try hierarchical clustering.

Classification and Clustering are separate things.  K-Means is the baseline method for clustering and it has many limitations , this is a great paper to start with

Jain, Anil K. "Data clustering: 50 years beyond K-means." Pattern recognition letters 31.8 (2010): 651-666.

Classification SVM is one of the most robust classifiers.

The SVM is a type of Supervised classifier and K-means is a clustering tool that is unsupervised. Both are very different from each other. During classification there is a set of features used as training set consisting of feature values with predefined class names and during testing classifiers tries to assign a class name to a new object with some features. But in case of clustering the unknown objects are divided into clusters depending on some relationship between them. You can not tell any classifier or clustering technique is better or best. It all depends on the type of data and features that are used for feature extraction. The feature dimension is also a big factor in case of constraints like computational cost. 

For your reference you can read the paper suggested by Mr. Goswami before and also you can read the following:

Boris Mirkin, "Mathematical classification and clustering: From how to what and why", Classification, Data Analysis, and Data Highways Studies in Classification, Data Analysis, and Knowledge Organization, pp 172-181, 1998.

thanks to all of you sir....

Agree Ludovic Journaux's answer

I agree with Chayan Halder.

You can not tell any classifier or clustering technique is better or best.

It all depends on the type of data and features that are used for feature extraction. The feature dimension is also a big factor in case of constraints like computational cost.

See: 

Jiawei Han, Micheline Kamber, and Jian Pei, Data Mining: Concepts and Techniques. Morgan Kaufmann, 3rd ed. , 2011

For classification you will use k-nearest neighbor instead of k-means. It works pretty well for sparse high dimensional data. You can even combine k-means and SVM. If you are interested may be you want to take a look to an ICANN 2002 paper on clustering with SVMs you can find in my publication list.

I thiank the  svm and k-means each has its advantages and disadvantages in  classification and clustering. And you can combine them too.

It depends on what kind of data do you want to cluster or to classify. I agree with Ludovic journaux. 

For unsupervised, you could use algorithms of Galois lattice generation, if your data is reprented as a binary relation.

You could also use random forest method for supervised classification.

Yes, all the above answers are correct. it's all depends on your application.

You should know what is your input or initial knowledge.

Next, you must concentrate on what you exactly want the method to bring or do for you.

If you want to have an automatic clustering you can try K-means or fuzzy c-means clustering methods with just determining the number of clusters you want. and if you have not to determine also the number of cluster center by yourself you can try AHFCM method wich produces initial optimise cluster number for fuzzy c-means algorithm.

On the other hand, if you want to do a classification and use reference data, you can use SVM method of classification.

Svm is very good algorithm

I think it depend on application. SVM is an accurate algorithm but needs an expert knowledge of the training sites .

I agree with Glauco da Silva .

For AI algorithm selection, I suggest to take a look at following cheat sheet that is provided by Sklearn:

https://scikit-learn.org/stable/tutorial/machine_learning_map/index.html

or another cheat-sheet such as:

https://twitter.com/KirkDBorne/status/1249725146509848579/photo/1

I like this resource (link) about clustering so much. Maybe it is all said by now, but it still can be interesting for someone reading this question :) >>> https://towardsdatascience.com/the-5-clustering-algorithms-data-scientists-need-to-know-a36d136ef68

It depends on the problem that want to be solved.

Absolutely, I have the same answer that Al-Khamees