Gene prediction, a fundamental task in genomics, plays a pivotal role in decipheringthe functional elements of genomes. However, accurate gene prediction remains a challenging endeavor, especially when dealing with non-model organisms that often lack extensive phylogenetically similar training data. Current methods which produce highly accurate results on small samples of DNA from model organisms routinely fail to achieve the same level of accuracy in non-model organisms. This dissertation presents two innovative ab-initio gene prediction methods based on neural networks and random forests, addressing the need for improved accuracy and reduced data requirements. Accurate predictions of coding sequences across whole chromosomes will be demonstrated by these methods even when trained on small amounts of data from un-related organisms. Prediction accuracy will also be compared to other lead- ing methods on a highly diverse range of non-model organisms. It will be shown that these methods not only reduce the reliance on extensive training data but also ele- vate the accuracy of gene prediction, outperforming several existing techniques. As genomics continues to expand its scope to encompass a broader array of species, the contribution of these methods holds promise in advancing our understanding of the genetic landscapes of non-model organisms.