Institution profile

Curtin University

Academic institutionaustralasia · au
Official website
Research library93linked papers
Opportunities0open roles
Selected work

Representative Papers

A deep learning and machine learning approach to predict neonatal death in the context of São Paulo

Mar 01, 2024International Journal of Public Health Science (IJPHS)

This study addresses neonatal mortality risk prediction in the São Paulo region using a large-scale, real-world birth dataset comprising 1.4 million records. To enable early identification of high-risk newborns, we establish a multi-model comparative framework and—novelty for this region—introduce Long Short-Term Memory (LSTM) networks for neonatal mortality prediction, overcoming performance limitations of conventional machine learning approaches. Experimental results demonstrate that the LSTM model achieves 99% accuracy, substantially outperforming baseline models including logistic regression, k-nearest neighbors (KNN), random forest (94%), XGBoost (94%), and convolutional neural networks (CNN). The findings empirically validate the efficacy of temporal modeling for perinatal risk prediction and yield a clinically deployable, high-accuracy early-warning system. This work establishes a new paradigm for neonatal mortality intervention in resource-constrained settings.

4 citationsRead paper

ANUBIS: Skeleton Action Recognition Dataset, Review, and Benchmark

May 04, 2022arXiv.org

Existing 3D skeleton-based action recognition research suffers from fragmented representation taxonomies and evaluation protocols misaligned with real-world scenarios; moreover, mainstream datasets lack critical dimensions—including rear-view perspectives, multi-person interactions, fine-grained or violent actions, and pandemic-era behaviors. To address these gaps, we propose a four-dimensional taxonomy (dataset design, spatial modeling, temporal modeling, and signal enhancement) and introduce ANUBIS: the first large-scale, multi-view 3D skeleton dataset explicitly designed for realistic challenges. ANUBIS features rear-view captures, 101 action classes (including 21 pandemic-related behaviors), and standardized recordings from 128 participants using Azure Kinect’s multi-sensor fusion. We further establish a unified benchmark framework, enabling reproducible evaluation of 12 state-of-the-art models. Our analysis identifies temporal modeling capacity and signal robustness as the primary bottlenecks limiting current performance.

4 citationsRead paper
Recent publications

Latest Papers

Designing Sustainable Federated Learning as a Service using Neural Architecture Search

Aug 14, 2026

This study addresses the challenges of infeasible user participation and training instability caused by carbon constraints in Federated Learning as a Service (FLaaS). We propose SFLaaS, a novel framework that innovatively constructs a demand-driven search space and a carbon feasibility assessment mechanism. Integrated with an adaptive scheduling strategy, this approach jointly optimizes model performance and user participation under hard carbon constraints. Experimental results demonstrate that SFLaaS effectively ensures federated training stability under heterogeneous sustainability constraints. Furthermore, it achieves simultaneous improvements in carbon compliance and model accuracy, establishing a new paradigm for green federated learning.

0 citationsRead paper

LoDA: A Level of Detection Aware Method and a Multimodal Sensing Benchmark for Object Level Change Detection

Aug 05, 2026

This study addresses the challenge of reliable object-level 3D change detection in multi-temporal urban LiDAR data, where existing methods suffer from ambiguous detection limits and inconsistent labeling. To overcome these limitations, this work proposes the first object-oriented 3D change detection framework that explicitly incorporates the Level of Detection (LoD) concept. By decoupling registration, geometric, and semantic modules, the approach propagates pose uncertainty into spatial change LoDs, thereby enhancing cross-temporal correspondence stability and suppressing false positives. The method integrates multi-modal sensing (LiDAR/GNSS/IMU), geometry-driven object proxies, regularized semantic and instance segmentation, and multi-dimensional features—including height, volume, and normal displacement—to produce five confidence-aware change labels. Evaluated on the newly introduced LoDA benchmark, it achieves 95.0% accuracy, 90.8% macro F1, and 83.0% macro IoU, and attains 96.81% average accuracy and 89.52% average change IoU on Urb3DCD-V2, significantly outperforming state-of-the-art approaches.

0 citationsRead paper