Text Categorization is problem assigning text documents into fixed number of pre-defined categories. Feature selection and Term weighting are two important steps that decide the result of any Text Categorization problem. In this paper we focus on two things first is to develop effective term weighting by proposing ne w term weighting scheme and second is to utilize the parallel and distributed processing capability of Hadoop MapReduce for training and testing of dataset. These two things leads to great performance improvement of text categorization by remarkable improvement in accuracy with a significant reduction of computational cost. Also because of the use of Hadoop MapReduce it reduces the training and testing time significantly.