Integrated multivariate and machine learning approaches reveal trait variability and selection priorities in tamarind (Tamarindus indica L.) germplasm of Karnataka, India

Tamarind ( Tamarindus indica L.) is a multipurpose perennial tree widely cultivated in the dry and semi-arid tropics, with significant nutritional, industrial and ecological importance. Despite its economic value, systematic characterization of tamarind germplasm remains limited. In this study, 59 accessions from Southern Karnataka were evaluated for morphological and nutritional traits using integrated statistical and machine learning approaches. Substantial variability was recorded in tree height (6.1–18.7 m), crown spread (4.3–15.2 m), pod weight (10.6–27.5 g), pulp weight (5.4–14.9 g), and ascorbic acid content (5.7–19.3 mg/100 g), reflecting wide diversity among accessions. Correlation analysis revealed strong associations between pod weight, pulp weight and pod circumference, indicating their central role in yield determination. Principal Component Analysis (PCA) explained 63.9% of total variance with the first three components, highlighting pod and pulp traits as major contributors. Cluster analysis grouped the accessions into four distinct clusters, each characterized by specific trait combinations, facilitating targeted parental selection. Random Forest feature importance ranked pod weight, pulp weight and crown size as the most discriminating traits, thereby prioritizing them for breeding programs. Cluster-wise trait profiling using heatmaps and radar charts provided clear visualization of inter-cluster differences, further supporting selection strategies. This integrated approach combining conventional, multivariate and machine learning methods offers a holistic framework for tamarind germplasm evaluation. The findings provide actionable insights for breeders to develop ideotypes with desirable canopy architecture, high pulp yield and superior nutritional quality, thereby strengthening conservation and improvement programs for tamarind.