Abstract:
This dissertation presents a formal approach for identification and specification based on the well-established object-oriented paradigm then employs artificial neural network (ANN) to classify software component repository into similar component cluster. This clustered component repository is subsequently indexed using non-hierarchical and hierarchical indexing based on three unsupervised neural network techniques, namely, Rival Penalized Competitive Learning (RPCL), Fuzzy Subtractive Clustering (FSC), and Kohonen's Self-Organizing feature Map (SOM). This step is referred to as coarse grain classification. In this study, analysis of the proposed the approach has been conducted to measure their efficiency in terms of precision, recall, and training time. The results confirmed that Rival Penalized Competitive Learning (RPCL) was the superior technique. Subsequent retrieval of software component belonging to the cluster partition whose center is closest to the requirements can thus be retrieved and participated in selecting the most suitable software component at the fine grain level. Consequently, this approach not only is suitable for multidimensional data, but also furnishes a basis for machine learning applications.