Cracking the Genetic Codes: Exploring DNA Sequence Classification with Machine Learning Algorithms and Voting Ensemble Strategies

In the domain of bioinformatics, DNA sequence classification is an indispensable tool that spans various scientific disciplines, contributing to scientists' understanding of biology, aiding in the identification of genes, regulatory elements, and the functional significance of distinct genomic regions. Moreover, it plays a vital role in disease diagnosis, treatment strategies, drug discovery, evolution, agriculture, forensic identification, environmental monitoring and more. The classification process involves the intricate mapping of DNA sequences to distinct classes based on the arrangement of nucleotides. A fractional mutation in the sequence corresponds to a nuanced shift in the assigned class. Every numerical instance, serving as a depiction of a particular class, is closely associated with a specific gene lineage. In this study, for the DNA sequence preprocessing, both K-mer counting and count vectorization were used respectively. Afterwards, we utilized a variety of classifier models, encompassing Multinomial naive bayes (MNB), Logistic regression (LR), Random forest (RF), LightGBM (LGMB), XGBoost (XGB), K-nearest neighbors (KNN) and Decision tree (DT) algorithm on three types of DNA sequence datasets (Human, Chimpanzee & Dog) to identify each of sequence's corresponding gene class (0, 1, 2, 3, 4, 5, & 6). Then, the highest three and highest five classifier models were picked based on their accuracy scores. Afterwards, both soft voting and hard voting ensemble methods were implemented on this cluster of fundamental models to effectively leverage their collective predictive strength. The soft voting ensemble on the best three models consistently reached the highest accuracy across all three datasets. Employing this ensemble method, the human, chimpanzee, and dog datasets exhibited highest performance metrics i.e. accuracy, precision, recall, and fl-scores of (98.42 %, 98.41 %, 98.40%, 98.40%), (92.28%, 92.40%, 92.30%, 92.10%), and (70.12%, 73.10%, 70.10%, 69.20%) respectively.