Approaching Software Cost Estimation Using an Entropy-Based Fuzzy k-Modes Clustering Algorithm

A new software cost estimation approach is proposed in this paper, which attempts to cluster empirical, non-homogenous project data samples via an entropy-based fuzzy k-modes clustering algorithm. The target is to identify groups of projects sharing similar characteristics in terms of cost attributes or descriptors, and utilise this grouping information to provide estimations of the effort needed for a new project that is classified in a certain group. The effort estimates produced address the uncertainty and fuzziness of the clustering process by yielding interval predictions based on the mean and standard deviation of the samples having strong membership within a cluster. Empirical validation of the proposed methodology was conducted using a filtered version of the ISBSG dataset and yielded encouraging results both in terms of practical usage of the clustered groups and of approximating effectively project costs.