Account compromising is a leading security chal-lenge to Online social networks. The compromised Online SocialNetwork accounts are used for malicious activities. The challengesposed by compromised accounts are more severe than many othersecurity attacks because because once an account is compromised,the attacker gains complete access to the account. The attackersmake use of the well established trusted network associated withthe compromised accounts. One of the challenges in detectingcompromised twitter account is the uncertainty inherent to thetweets. Another challenge is the complexity in user behaviourexhibited in various attributes related to the user account andtweets.Existing works found in literature fails to address theuncertainity in tweets. The aim of this paper is to proposesan approach to detect compromised twitter accounts usinga combined approach of machine learning and multi-criteriadecision making. In order to address the uncertaininy of tweetscognitive clound model is applied. New features are derived in thiswork to represent tweets.This proposed approach exhibits betteraccuracy compared to Weighted Sum method and WeightedProduct method comes under the multi criteria decision makingmethods and with random forest classifier.