Dynamic modeling based on machine learning algorithms for unstructured web data pattern recognition and predictions
Vincent Andrew Akpan, Joshua Babatunde Agbogun, Bamidele Ayodeji Oluwade · Journal of ICT Development Applications and Research · 2025
Extraction of structured information and/or data from the huge semi-structured and unstructured webpages available on the Internet is a challenging multidisciplinary field which demands advanced algorithms for data extraction, modeling, pattern recognition and prediction. This paper presents a dynamic modeling algorithm based on machine learning for unstructured web data pattern recognition and prediction. The machine learning approach is formulated in the context of an online neural network-based adaptive recursive least squares (ARLS) algorithm with an adaptive exponential forgetting and resetting parameter. The proposed machine learning algorithm based on ARLS is validated using the Kogi State University (KSU) website as a case study and its pattern recognition and prediction performances are compared against an error backpropagation with momentum (EBPM) algorithm. Five case studies from the KSU website have been considered in this paper and the proposed neural network-based ARLS shows superior performances compared to the EBPM in terms of fast convergence and small predictions errors. The proposed algorithm can be adapted for online information extraction of structured web data from unstructured webpages.