Online social media websites like Twitter has become one of the most popular platforms for people to obtain or spread information. However, in absence of any moderation and use of crowd sourcing, there is no guarantee that the information shared is credible or not. This makes online social media highly susceptible to the spread of rumors. As part of our work, we investigate in retrospect a dataset on which rumor detection was done in the past in 2009 and perform machine learning algorithms like k-nearest neighbor and naive bayes classifier to detect tweets spreading rumors. We present the results of our retrospective analysis and extraction of user attributes. An algorithm for preprocessing on tweet content is proposed to retain key information to be passed on to learning algorithm to obtain improved results as far as rumor detection accuracy is concerned.