<![CDATA[This study compares the performance of three clustering algorithms, namely, K-means, HDBSCAN, and Gower distance-based hierarchical clustering, for identifying latent profile groups in ASD data. The novelty of this research lies in the integrated evaluation of conventional numerical clustering methods and a mixed-data clustering framework that incorporates clinically relevant categorical variables, enabling more interpretable ASD subgroup discovery. A quantitative experiment was conducted using a validated dataset of 500 ASD records comprising demographic attributes, symptom indicators, and developmental assessment scores (24 features). K-means and HDBSCAN were applied to eight numerical profile dimensions, whereas hierarchical clustering used 18 mixed numerical–categorical features with Gower distance. Cluster quality was evaluated using the silhouette score, Calinski–Harabasz index, and density-based clustering validation (DBCV). The results indicate that K-means achieved the strongest global partition structure (silhouette = 0.1775; Calinski–Harabasz index [CH] = 59.54), outperforming the other methods as a practical baseline for structured numerical ASD data. HDBSCAN showed competitive clustering performance while uniquely identifying 19.8% of observations as noise, suggesting its usefulness for detecting atypical or rare ASD cases. Although hierarchical–Gower produced lower internal metrics (silhouette = 0.0441), it successfully integrated categorical clinical variables, offering richer contextual segmentation than purely numerical approaches. These findings demonstrate that no single algorithm is universally optimal for ASD profiling; instead, clustering selection should align with analytical objectives, such as compact grouping, anomaly detection, or clinically interpretable subgrouping.]]>
