Breast cancer is the most common cancer globally and often diagnosed at advanced stages in Indonesia. Metabolomic profiling has emerged as a promising approach for identifying biomarkers associated with breast cancer (BC). However, the specificity and clinical applicability of candidate metabolites remain under investigation. This study investigates untargeted plasma metabolomic profiles of breast cancer patients to find candidate metabolite signatures of breast cancer. Plasma samples from 24 breast cancer patients and 24 healthy controls (HC) were analyzed using untargeted Gas Chromatography-Mass Spectrometry (GC-MS). A machine learning (ML) approach was utilized to validate the metabolites. Differential metabolites were identified and analyzed to explore altered metabolic pathways associated with BC. Several metabolites, including D-glucose, citric acid, lactic acid, L-hydroxyproline, and glutamic acid, were significantly different between BC and HC groups. Those metabolites correlated with arginine/proline metabolism, glycolysis, and alanine/aspartate/glutamate pathways. ML validation yielded favorable results for these metabolites as candidate metabolite signatures of breast cancer (AUC > 0.8, accuracy > 80%). Further subset analysis showed reduced dihydrouracil in late stage. Untargeted plasma metabolomic analysis combined with machine learning effectively identified a potential candidate metabolite signature for breast cancer. These findings improve understanding of breast cancer metabolic alterations and highlight promising pathways for early diagnosis. Nevertheless, further validation in larger, well-controlled studies is required to establish their diagnostic utility. © 2026 by the authors.