Abstract
Flowering time is an important developmental stage in plants, influenced by multiple genes and environmental factors. Understanding its genetic basis and interaction with the environment facilitates the development of improved varieties adapted to different environments. Conventional Genome-Wide Association Studies (GWAS) have been widely used to associate genetic markers with heritable traits, but they do not inherently capture interactions among single nucleotide polymorphisms (SNPs) or between SNPs and the environment. Machine Learning (ML) approaches can model these interactions and improve trait prediction even in the presence of noise and missing data. In this study, multi-environment lentil (Lens culinaris Medik.) data were analysed using GWAS and two widely used ML models, Random Forest and XGBoost, to identify genetic markers associated with flowering time. Model interpretability was enhanced using Explainable AI (XAI) techniques, including SHapley Additive exPlanations. GWAS identified eight significant loci across chromosomes one, two, five and seven, with the most significant SNP located at Chr2_530433205, while ML approaches identified nine markers on chromosomes one, two, three, five and seven, with the most significant SNP at Chr7_523220088. The majority of the identified markers were linked to candidate genes for flowering, while ML also identified potential epistasis. These findings highlight ML as a powerful complementary tool to GWAS for trait association.