
Research Article
Archiving 4.0: Dataset Generation and Facial Recognition of DRC Political Figures Using Machine Learning
@INPROCEEDINGS{10.1007/978-3-031-35883-8_4, author={Ferdinand Kahenga Ngongo and Antoine Bagula and Olasupo Ajayi}, title={Archiving 4.0: Dataset Generation and Facial Recognition of DRC Political Figures Using Machine Learning}, proceedings={Emerging Technologies for Developing Countries. 5th EAI International Conference, AFRICATEK 2022, Bloemfontein, South Africa, December 5-7, 2022, Proceedings}, proceedings_a={AFRICATEK}, year={2023}, month={7}, keywords={Archiving Deep Learning Democratic Republic of Congo Transfer learning Face recognition}, doi={10.1007/978-3-031-35883-8_4} }
- Ferdinand Kahenga Ngongo
Antoine Bagula
Olasupo Ajayi
Year: 2023
Archiving 4.0: Dataset Generation and Facial Recognition of DRC Political Figures Using Machine Learning
AFRICATEK
Springer
DOI: 10.1007/978-3-031-35883-8_4
Abstract
It is widely recognized that digital archiving represents many advantages compared to manually preserved documents. These include i) reduced risk of losing data, ii) eco-friendliness, iii) data security, iv) faster access to data, v) simple data management, vi) overall costs saving, and vii) potential for data recovery. However, in contrast to developed nations, which have experienced a steady maturity in the field, digital archiving is still in its infancy in developing countries, especially in Africa, where years of slavery, colonization, both economic and political issues, and wars have deprived nations of their history. Building around techniques of the fourth industrial revolution (4IR), this paper addresses this gap by proposing a digital archiving model called “Archiving 4.0”. The model curates a dataset of images of political figures in the Democratic Republic of Congo (DRC) from different sources and classifies these images using machine learning techniques to achieve facial recognition that will help recognize historical people in photographs and videos. To the best of our knowledge, the proposed dataset constitutes the first attempt at digital archiving in the political space of the DRC and provides an example that can be emulated to spin out related works on the African continent where digital archiving is needed for political research studies and also for the preservation of the history of nations. Through performance evaluation, accuracy, precision, recall, and loss; the paper reveals that Transfer learning outperforms traditional machine learning on different metrics of interest when using the generated dataset: 93% of validation accuracy.