Neoplasm classification and prediction is an important and challenging task. The main objective of this study was to propose an enhanced approach with features of data segmentation, data clustering and analysis in terms of cost factor and cross validation. The clinical data was collected and partitioned in some fixed size segments. Each segment was further decomposed in variant sized clusters (larger and smaller) with application of cleansing data filter for specific data conversion and translation. These clusters were then passed through different combinations of network architectures. The selection of efficient architecture was based on degree of achievement for neoplasm classification (malignant and benign) and root mean square values (smaller value depicts feasible network). This study demonstrated an enhanced approach over granular diverse data clusters and elaborated the results with graphical simulation in differentials of active and cross validation cost factors against percentage of achievement. The classification and prediction was 100% for benign and 99.05% for malignant masses with least root mean square value of 0.017. The novelty of the proposed approach is that it demonstrated results at high degree of granularity and was quite suitable for analysis of any problem size. Overall, it provided an opportunity for further work to outperform as an enhanced scalable approach for neoplasm classification and prediction with neural network.