This paper proposes a variant and an improved attribute chosen framework of Dynamic Relevance and Joint Mutual Information Maximization (DRJMIM) attribute chosen method. DRJMIM-IW, just like DRJMIM, is also based on maximum of minimum criterion for feature relevancy to distinguish between redundant and interactive attributes but differs in the way the weight of a candidate feature is calculated. Information theoretic based weight measure IW (Interaction Weight) between a potential feature and chosen feature is used as a measure for dynamic updation of the relevance term. The joint mutual information between a potential feature, chosen feature and target are weighted to distinguish between interactive and redundant features. We evaluated our proposed method on Cervical cancer risk factor data downloaded from UCI repository. With SVM, 5NN, CART, Naive Bayes and Random forest classifier as learning models and accuracy as evaluation metrics, we performed the experiment. The proposed method outperforms the base algorithm DRJMIM in almost all cases.