30 August 2021 in IEEE Access
Most of the established neural network architectures in computer vision are essentially composed of the same building blocks (e.g., convolutional, normalization, regularization, pooling layers, etc.), with their main difference being the connectivity of these components within the architecture and not the components themselves. In this paper we propose a generalization of the traditional average pooling operator. Based on the requirements of efficiency (to provide information without repetition), equivalence (to be able to produce the same output as average pooling) and extendability (to provide a natural way of obtaining novel information), we arrive at a formulation that generalizes average pooling using the Zernike moments. Experimental results on Cifar 10 , Cifar 100 and Rotated MNIST data-sets showed that the proposed method was able to outperform the two baseline approaches, global average pooling and average pooling $2 \times 2$ , as well as the two variants of Stochastic pooling and AlphaMEX in every case. A worst-case performance analysis on Cifar-100 showed that significant gains in classification accuracy can be realised with only a modest 10% increase in training time.
4 November 2024 in IEEE Access
This work presents a framework for predicting the likelihood of terrorist incidents within a specific time period in a target country based on machine learning and deep learning models. These models are trained on localised news data sourced from the Global Event, Language, and Tone Database (GDELT) Project for the target country, while the terrorist incidents recorded in the Global Terrorist (GTD) Database serve as the ground-truth. To cope with the massive volume of information acquired from localised news, a novel feature selection process based on Random Forest models is proposed, where the predictive power of each feature is calculated and only features estimated to have sufficient predictive power are subsequently considered. Experiments are performed by considering the United Kingdom as a case study and six predictive models are evaluated across four evaluation metrics. The experimental evaluation results indicate that the proposed framework enhances the predictive power of both machine learning and deep learning models that use the proposed feature selection in forecasting the likelihood of terrorist incidents, surpassing models trained (i) on all baseline features and (ii) on features selected by two well established feature selection approaches. Overall, the proposed framework aims at assisting counter-terrorism efforts by providing a projection of the likelihood of such terrorist incidents and thus the overall terrorism risk in a target country, using the most valuable features provided by GDELT.
26 October 2026 in Expert Systems
A tender receiving only one offer (single bidding) is generally recognized as a sign of restricted competition and a potential corruption risk indicator. Existing corruption risk models rely predominantly on ex post information, limiting their usefulness for early stage prediction. This paper introduces a network‐enhanced supervised learning framework that predicts single bidding outcomes at the time of tender announcement, using only features available at the first call for tender. The framework integrates conventional procurement attributes with historical buyer behaviour metrics and with relational features derived from annual buyer–supplier interaction networks, including centrality measures, community structure and node embeddings. To ensure temporal validity, all features are constructed exclusively from data available prior to tender publication. The methodology is evaluated across three European countries, Bulgaria, Croatia and Serbia, using country‐specific models to account for institutional and market heterogeneity. Results show that the Random Forest model enriched with network‐based and historical features consistently outperforms baselines relying solely on transactional attributes, achieving macro F1‐scores between and . These results demonstrate that single bidding can be forecast reliably at the time of tender announcement and highlight the value of incorporating relational structures and buyer behaviour into procurement analytics. The proposed framework offers a scalable foundation for early warning systems that support proactive integrity monitoring and more transparent procurement governance.