Efficient strategies are necessary to encode DNA sequence characteristics in a suitable format for DNA motif discovery or DNA classification tasks in deep learning neural network models. The selection of an encoding scheme is based upon the desired form of representation, whether it be a vector or an n-dimensional matrix, as well as the particular problem at hand and the underlying architecture of the deep learning model being employed. Convolutional neural networks (CNNs) often utilise a grid-based representation by employing either one-hot encoding or k-mers to represent a given genome or protein sequence. The utilisation of one-hot encoding and label encoding is highly valuable in the context of Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models, as the precise positioning of characters at each time step plays a critical role. In order to capture intricate properties, it is possible to convert characters into vectors, which can then be employed as embeddings. Thus, the objective of this study is to conduct a comparative analysis of four distinct feature encoding algorithms for genomic sequences, implemented within the framework of a convolutional neural network. In order to streamline the procedure, we employ nucleobase sequences of predetermined length as the input data. The results of this study indicate that word embeddings exhibit superior performance compared to alternative encoding techniques in the context of genomic data analysis.