A reproducible entropy-weighted Gower-CLARA-TOPSIS procedure for mixed-type MSME segmentation and prioritization

In this paper a reproducible approach for segmentation of mixed-type data and the within-cluster prioritizing is presented. The procedure keeps raw categorical values for Gower dissimilarity, binary or count representations only for entropy weighting, applies CLARA to repeated samples with PAM medoids, and evaluates weighted and Top-N feature configurations with the Silhouette Coefficient, medoid-based Davies-Bouldin Index, and Calinski-Harabasz Index. The chosen configuration is then utilized to rank within clusters, instead of between clusters using TOPSIS to evaluate the economic-operational preparedness of similar firms. It presents a worked example with 1049 de-duplicated MSME records from Sampang, Indonesia, with numerical, binary, count and categorical attributes. The optimum configuration selected four operational and financial aspects and increased the Silhouette Coefficient from 0.4183 to 0.5588, decreased the Davies-Bouldin Index from 1.2393 to 0.6181, and boosted the Calinski-Harabasz Index from 357.7787 to 992.3958. By combining entropy-based feature weighting, weighted Gower dissimilarity, CLARA sampling, and Top-N feature selection, the procedure provides transparent mixed-data segmentation while preserving categorical semantics and separating clustering from peer-group-based TOPSIS ranking. A reference Jupyter Notebook implementation and an anonymized or de-identified demonstration dataset are supplied to support inspection and reuse of the workflow.