Enhancing Biogeographical Ancestry Prediction with Deep Learning: A Long Short-Term Memory Approach

Nowadays, human biogeographical ancestry prediction plays an important role in many domains, such as the forensic domain, to detect missing or suspected people. Despite the advantage and capability of these deep learning models, there were limited investigations on identifying human biogeographical ancestry using deep learning approaches. In this research, we propose to predict biogeographical ancestry using a deep learning approach to distinguish between seven populations (Africans, Europeans, Central-South Asians, Middle-East Asians, East Asians, Native Americans, and Oceanians). We used the Long Short-Term Memory (LSTM) approach to enhance the overall current accuracy models, especially for populations that have gene similarity such as (Europeans, Middle-East Asians, and Central-South Asians). We employed a stratified K-fold cross-validation technique to prevent overfitting and ensure an equal distribution of samples for each fold. The results showed that our model outperformed the existing deep learning algorithm Convolutional Neural Network (CNN), by achieving an overall accuracy of 90.88.