Age-Driven Clustering of Cancer Types: A Data-Driven Taxonomy Using Machine Learning

Abstract

Global mortality remains heavily driven by cancer, with patient age serving as a primary determinant of both incidence and pathological distribution. This study evaluates the predictive weight of age on cancer typology through an unsupervised machine learning lens, utilising a dataset of over 5,000 de-identified records from the National Oncology Centre in Yemen. The authors constructed a Python-based analytical framework to handle data preprocessing and imputation, subsequently comparing the efficacy of K-Means, Agglomerative Clustering, and Gaussian Mixture Modelling (GMM). GMM proved the most robust approach, yielding a Silhouette Score of 0.6135. Consequently, this model formed the basis for the final analysis. To ensure the clusters reflected genuine biological trends rather than stochastic noise, the authors validated the results using ANOVA and Chi-Squared tests. The analysis identified two distinct, age-stratified cohorts. The first, encompassing patients aged 2–40, showed a higher prevalence of bone marrow, lymphatic, thyroid, and breast malignancies. In contrast, the older cohort (ages 41–101) was characterised largely by breast and gastrointestinal cancers. These results establish age as a measurable, objective predictor of cancer distribution. Such findings suggest that clinical screening protocols and diagnostic priorities must be calibrated more closely to specific age demographics to enhance early detection efforts.