Nocardiosis is a major problem affecting fish that are farmed in seacages as well as freshwater fish; therefore, deciphering the bacteriological features of Nocardia seriolae is crucial. In particular, a number of studies over the past two years have reported the genome sequence of N. seriolae, and a comparative genomics approach is expected to yield valuable information on its epidemiological characteristics. The purpose of this study was to perform whole-genome sequence analysis of N. seriolae MH196537 from the Japanese eel and to investigate the significant differences noted between strains isolated from freshwater fish and marine fish by using Random Forest, a reliable machine learning algorithm. The Pacbio platform was employed to sequence the MH196537 strain, and genomic information from the other 16 strains was used for comparative analyses. All coding sequences of the 17 strains were categorized in RASTtk Sub-systems. The MH196537 strain had one contig, and it shared a high average nucleotide identity (ANI) with the freshwater strains (0.9994 - 0.9999) rather than the seawater strains (0.9985 - 0.9994). Moreover, 22 RASTtk subsystems carried a different number of genes from each N. seriolae. The fatty acids, lipids, and isoprenoids subsystem showed the highest mean decrease in the Gini index of over 1.5. Interestingly, freshwater strains were found to harbor all of the genes for both the mevalonate (MVA) and non-mevalonate pathways (MEP), whereas only the MEP existed in strains from diseased marine fish. Considering the differences in the byproducts of isoprenoids from the different pathways, it is likely that this will affect host-pathogen interactions; therefore, harboring the different pathways for the synthesis of isoprenoids could be an important pathogenic factor of N. seriolae.