Thank you very much Pradeepa, for your advice. Can you share your research paper (published) and other literature that you refer to study .

For your question: I suggest you to study bit of python and use packages which lead us to think plenty of ways to proceed the researches for Data mining.

Kappa Statistics is used to measure the inter-rater agreement for categorical datasets.

If you want to implement kappa for your evaluation: 

In Weka you can follow this link to get an idea for implementation:

http://nullege.com/codes/search/weka.classifiers.Evaluation.kappa

In Python:

you can use scikit-learn library.

from sklearn.metrics import cohen_kappa_score, make_scorer

kappa = cohen_kappa_score(y_true, y_pred, labels=None)

More information is available here:

http://scikit-learn.org/stable/modules/generated/sklearn.metrics.cohen_kappa_score.html

http://nullege.com/codes/search/weka.classifiers.Evaluation.kappa

http://scikit-learn.org/stable/modules/generated/sklearn.metrics.cohen_kappa_score.html

Cohen's kappa coefficient used for inter-rater agreement. This evaluaton we use if we have several measurements from raters or judges and we would like to know how identic or agree those measurement that come from several raters or judges.

In your task, I think it will be more suitable if you use Accuracy, Recall, Precision, and AUC (area under curve). Rapidminer, weka, etc provide those evaluation. 

Thanks, MR.Hiran Wijesinghe, Mr.Amit Tyagi and Mr.Adi Wijaya for your suggestions.

It helping me lot. once again thank you

Kappa Statistic and Mean Absolute Error (MAE) are very important to determining and judge the best model in your analysis.

Please, see this publication for full information:

https://www.researchgate.net/publication/318338750_Supervised_Machine_Learning_Algorithms_Classification_and_Comparison?_iepl%5BviewId%5D=FY75L7KN1r7iYF6EqX57NFx0sUS96keeqFhb&_iepl%5Bcontexts%5D%5B0%5D=prfhpi&_iepl%5Bdata%5D%5BstandardItemCount%5D=0&_iepl%5Bdata%5D%5BuserSelectedItemCount%5D=1&_iepl%5Bdata%5D%5BtopHighlightCount%5D=1&_iepl%5Bdata%5D%5BtopHighlightIndex%5D=1&_iepl%5Bdata%5D%5BfeaturedItem1of1%5D=1&_iepl%5BtargetEntityId%5D=PB%3A318338750&_iepl%5BinteractionType%5D=publicationTitle

 You need to check the value of Kappa Statistic against MAE, where Kappa Statistic is greater than MAE, then the model is reliable for prediction.

Article Supervised Machine Learning Algorithms: Classification and Comparison

Can i do kappa statistics metric for find the best model among 3 different models?

YES you can use Kappa Statistic to determine the best model among 3 different models.

Kappa Statistic is a metric that compares an observed accuracy with an expected accuracy (Random Chance)