Disease prediction and classification using medical record datasets is a challenging data mining research problem that also requires attribute value imputation to be carried implicitly. Medical datasets that are available in public databases are not free from missing values and this is also true when data is collected and sampled through various clinical trials. In this context, there is always a need to turn up with new approaches and methods for accurate and efficient analysis of medical records. Several imputation strategies are proposed in the literature and each of them have reported accuracies achieved on benchmark datasets. However, a better imputation approach always helps in improving classification accuracies and this in turn helps to more accurate disease prediction . An approach for imputing medical records is proposed in this paper. We name the approach as Class -Based-Clustering-Imputation (CBC-IM). Experiments are carried out on several benchmark datasets. Results achieved using our imputation approach is compared to existing imputation approaches using classifiers such as KNN, SVM and C4.5. The results show improved performance on most of the datasets and are almost nearer to remaining approaches discussed in this paper.