Application of Clustering Techniques for the Analysis and Classification of Firms in the Automotive Industry

According to this study, five clustering algorithms are compared theoretically and experimentally. The clustering algorithms are DBSCAN, KMeans, Spectral Clustering, OPTICS, and K-Medoids. By using 20 automotive companies’ financial data, we evaluate the performance of each algorithm based on intra- and inter-cluster variance, noise robustness, and data perturbation sensitivity. The results demonstrate that the K-Means technique demonstrates the highest stability towards data variation, while the DBSCAN and OPTICS techniques are able to handle noise and discover clusters of different arbitrary shapes. Although spectral clustering is capable of capturing nonlinear structures, it is more sensitive to parameter changes. K-Medoids uses real data points as cluster centers, thereby providing robustness to outliers. One of the novel contributions of this work is the sensitivity analysis, which highlights the method’s robustness or fragility under controlled perturbations of the data. Each of these deductions is useful in choosing clustering methods according to the application.