Abstract Introduction: Epigenetic changes including methylation occur early in tumorigenesis and are known to be highly pervasive across a tumor type. The DNA methylation pattern can vary between individuals with or without cancer and also in patients with cancer of different origins. In this study, we investigated the discovery of comprehensive markers and validation of the diagnostic and prognostic potential of colorectal cancer. Methods: Patients with various stage of colorectal cancer were eligible for the current study, mainly stage III. We generated genome-wide methylation data from 379 colorectal cancer (CRC) tissues and 330 available paired adjacent normal mucosa tissues from Korean patients by Illumina EPIC Human Methylation microarray targeting 860,000~ CpG sites. A machine learning algorithm was used to build an optimized prediction model and select the tumor specific markers based on through theses CpG sites. Then, the risk score was devised for prognosis using this marker set. Finally, in order to validate the rule of CpG sites, the genomic location and pathway enrichment analysis was performed by CHROMHMM and Metascape. Results: 379 colorectal cancer tissues and 330 paired adjacent normal mucosal tissues were included for analysis. Clinical stages of the included patients were, III in N (61.2%), II in N (19.5%), and IV in N (11.9%). A total of 305 methylation markers that showed statistically significant differences between normal and cancer tissues were selected. Our model could accurately identify CRC (areas under the curve for the training and validation cohorts: 0.968 and 0.984, respectively). Using our prediction model, the colorectal cancer patients were predicted as colorectal cancer accurately in the methylation data from TCGA (COREAD; colorectal cancer tissue DNA) and GEO dataset (plasma cfDNA from colorectal cancer patients). The risk score comprising the subset of 305 methylation markers was calculated, and poor prognosis was predicted in the high-risk score group (overall survival P <0.1, progression-free survival P < 0.01). Gene ontology (GO) enrichment analysis showed that the 305 CpG sites were enriched in transcription regulatory regions (160/305, 52.5%) and were associated with developmental process and carcinogenesis (GO: 0032502, P <0.001; C4721208, P <0.001). Conclusions: In summary, the performance of our prediction model with these 305 CpG sites was highly accurate for CRC diagnosis, and the optimized risk score could predict the prognosis of Korean CRC patients Citation Format: Jun-Kyu Kang, Su Yeon Kim, Yoojoo Lim, Hwang-Phill Kim, Tae-You Kim. Use of an optimized machine learning algorithm to develop DNA methylation markers for detecting colorectal cancer (CRC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2800.