Artificial Intelligence Research

This book constitutes the refereed proceedings of the 6th Southern African Conference on Artificial Intelligence Research, SACAIR 2025, held in Cape Town, South Africa, during December 15, 2025.

The 36 full papers presented in this volume were carefully reviewed and selected from 150 submissions.The papers are organized into the following topical sections: Algorithmic and Data Driven AI; Socio-technical and Human-Centred AI; and Responsible and Ethical AI.

Klappentext

.- Algorithmic and Data Driven AI. .- Which contextual topic modelling algorithm is the best?. .- Leveraging Language Models for Document Type Classification in Low-Resource Afrikaans Archives. .- Hybrid Automatic Modulation Classification for Increased Robustness under White-Box Adversarial Attacks. .- Counterfactual Explanation Model for Personalised Dietary Interventions in Anaemia Patients. .- Feature extraction for plant growth estimation. .- Exploring the impact of different loss functions for anomaly prediction in a mineral processing plant. .- Modifying class distributions to improve the classification of minority group examples in a class-imbalanced dataset. .- Evaluating OCR performance on food packaging labels in South Africa. .- Investigating the Impact of Multilingual Pre-trained Speech Models on Gender Bias in ASR for Low Resource African Languages. .- A Two Stage Pipeline for Automated Caries Detection on Single Tooth Images from Panoramic Radiographs. .- Exploring syllable similarity across South African languages through self-supervised speech representation. .- Data Pruning: Redundant, Problematic, and Interdependent Samples. .- Restoring Neural Network Plasticity for Faster Transfer Learning. .- AfroXLMR-Comet: Multilingual Knowledge Distillation with Attention Matching for Low-Resource Languages. .- Code-Switch Pretraining for Improved Cross-Lingual Alignment in Low-Resource Languages. .- Identification of Social Media Users that Perpetuate Xenophobic Attitudes and Hate Speech Narratives in South Africa. .- HAPI-FedProx: Heterogeneity-Aware Adaptive Proximal Optimization for Federated Learning. .- Fog-Based Deep Learning for Real-Time Cold Chain Temperature Prediction Using IoT Data. .- Robust Driver Distraction Recognition via Lightweight Body-Part Association and Object Context on NVIDIA Jetson. .- Nearest-Class Mean and Logits Agreement for Wildlife Open-Set Recognition. .- Semi-Supervised Object Segmentation via Active Learning for Efficient Ecological Monitoring. .- Scaling behavior of Encoder Language Models in Low-Resource Settings. .- Enhancing Target Re-Identification via Model Fusion and Knowledge Distillation of Pre-trained Foundation Models. .- Heart Disease Prediction: A Comparative Study of Optimizers’ Performance in Deep Neural Networks. .- Socio-technical and Human-Centred AI. .- AI-Enabled Cybersecurity Implementation: A Case Study of Critical Success Factors in a South African SOE. .- Large Language Model Evaluation: From Model-Centric Benchmarks to System-Level Metrics. .- Exploring Postgraduate Students’ Use and Perceptions of Generative AI as a Research Support Tool: A South African Case. .- An Awareness Framework for Sustainable Selection of LLMs in Business. .- Synthesizing Insights from AI in Africa: A Comprehensive Qualitative Analysis. .- Factors influencing the use of Autonomous Vehicles on Traditional Road Infrastructure in The Western Cape. .- An Agentic LLM-Powered Framework for Cholera Risk Prediction and Actionable Insights from Explainable ML and Statistical Insights. .- A Scoping Review on Generative AI prompting to optimize the workflow of healthcare professionals in sub-Saharan Africa. .- Detection Of Power Line Insulator Defects Using YOLOv10-N. .- Responsible and Ethical AI. .- Artificially Generated Minorities (AGMs): The Veneer of Algorithmic Bias Correction. .- Under-resourced and over-reseached: The ELSA approach to transdisciplinary AI research in low-trust neighborhoods. .- From Principles to Practice: Extracting Human-Centred AI Components from Governance Frameworks.

Inhalt

.- Algorithmic and Data Driven AI.

.- Which contextual topic modelling algorithm is the best?.

.- Leveraging Language Models for Document Type Classification in
Low-Resource Afrikaans Archives.

.- Hybrid Automatic Modulation Classification for Increased Robustness
under White-Box Adversarial Attacks.

.- Counterfactual Explanation Model for Personalised Dietary
Interventions in Anaemia Patients.

.- Feature extraction for plant growth estimation.

.- Exploring the impact of different loss functions for anomaly prediction
in a mineral processing plant.

.- Modifying class distributions to improve the classification of minority
group examples in a class-imbalanced dataset.

.- Evaluating OCR performance on food packaging labels in South Africa.

.- Investigating the Impact of Multilingual Pre-trained Speech Models on
Gender Bias in ASR for Low Resource African Languages.

.- A Two Stage Pipeline for Automated Caries Detection on Single Tooth
Images from Panoramic Radiographs.

.- Exploring syllable similarity across South African languages through
self-supervised speech representation.

.- Data Pruning: Redundant, Problematic, and Interdependent Samples.

.- Restoring Neural Network Plasticity for Faster Transfer Learning.

.- AfroXLMR-Comet: Multilingual Knowledge Distillation with Attention
Matching for Low-Resource Languages.

.- Code-Switch Pretraining for Improved Cross-Lingual Alignment in
Low-Resource Languages.

.- Identification of Social Media Users that Perpetuate Xenophobic
Attitudes and Hate Speech Narratives in South Africa.

.- HAPI-FedProx: Heterogeneity-Aware Adaptive Proximal Optimization
for Federated Learning.

.- Fog-Based Deep Learning for Real-Time Cold Chain Temperature
Prediction Using IoT Data.

.- Robust Driver Distraction Recognition via Lightweight Body-Part
Association and Object Context on NVIDIA Jetson.

.- Nearest-Class Mean and Logits Agreement for Wildlife Open-Set
Recognition.

.- Semi-Supervised Object Segmentation via Active Learning for Efficient
Ecological Monitoring.

.- Scaling behavior of Encoder Language Models in Low-Resource Settings.

.- Enhancing Target Re-Identification via Model Fusion and
Knowledge Distillation of Pre-trained Foundation Models.

.- Heart Disease Prediction: A Comparative Study of Optimizers'
Performance in Deep Neural Networks.

.- Socio-technical and Human-Centred AI.

.- AI-Enabled Cybersecurity Implementation: A Case Study of Critical
Success Factors in a South African SOE.

.- Large Language Model Evaluation: From Model-Centric Benchmarks
to System-Level Metrics.

.- Exploring Postgraduate Students' Use and Perceptions of Generative
AI as a Research Support Tool: A South African Case.

.- An Awareness Framework for Sustainable Selection of LLMs in Business.

.- Synthesizing Insights from AI in Africa: A Comprehensive Qualitative
Analysis.

.- Factors influencing the use of Autonomous Vehicles on Traditional
Road Infrastructure in The Western Cape.

.- An Agentic LLM-Powered Framework for Cholera Risk Prediction and
Actionable Insights from Explainable ML and Statistical Insights.

.- A Scoping Review on Generative AI prompting to optimize the
workflow of healthcare professionals in sub-Saharan Africa.

.- Detection Of Power Line Insulator Defects Using YOLOv10-N.

.- Responsible and Ethical AI.

.- Artificially Generated Minorities (AGMs): The Veneer of Algorithmic
Bias Correction.

.- Under-resourced and over-reseached: The ELSA approach to
transdisciplinary AI research in low-trust neighborhoods.

.- From Principles to Practice: Extracting Human-Centred AI
Components from Governance Frameworks.