Abstract: Objectives: Various diet scores have been established to measure overall diet quality, especially for the prevention of cardiovascular disease (CVD). Diet scores constructed by utilizing modern machine learning techniques may contain independent information and can provide better dietary recommendations in combination with the existing diet scores. Methods: We proposed a novel machine-learning diet quality score (DQS) and examined the performance of DQS in combination with the Healthy Eating Index-2015 (HEI2015), Mediterranean Diet Score (MED), Alternative Healthy Eating Index-2010 (AHEI) and Dietary Approaches to Stop Hypertension score (DASH score). The data used in this study were from the 2011–2012 to 2017–2018 cycles of the US National Health and Nutrition Examination Survey (NHANES). Participants aged above 20 self-reported their food intake and information on relevant covariates. We used an elastic-net penalty regression model to select important food features and used a generalized linear regression model to estimate odds ratios (ORs) and 95% CIs after controlling for age, sex, and other relevant covariates. Results: A total of 16756 participants were included in the analysis. DQS was significantly associated with coronary artery disease (CAD) risk after adjusting for one of the other common diet scores. The ORs for DQS combined with the HEI2015, MED, AHEI, and DASH scores were all approximately 0.900, with p values smaller than 0.05. The OR for DQS in the full score model including all other scores was 0.905 (95% CI, 0.828–0.989, p=0.028). Only marginal associations were found between DQS and other CVDs after adjusting for other diet scores. Conclusions: Based on data from four continuous cycles of the NHANES, higher DQS was found to be consistently associated with a lower risk of CAD. The DQS captured unique predictive information independent of the existing diet scores and thus can be used as a complementary scoring system to further improve dietary recommendations for CAD patients.