Abdul Aziz Hulleck
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Abdul Aziz Hulleck has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
- Kinda Khalaf
Khalifa University
15 shared publications - Marwan El-Rich
Khalifa University of Science and Technology
13 shared publications - Rateb Katmah
Department of Biomedical Engineering and Biotechnology, Khalifa University, Abu Dhabi, United Arab Emirates
11 shared publications - Muhammad Abdullah
Khalifa University of Science and Technology
10 shared publications - Aamna AlShehhi
Department of Electrical Engineering Khalifa University Abu Dhabi United Arab Emirates
7 shared publications - Abdelsalam Tareq Alkhalaileh
Khalifa University, Abu Dhabi, United Arab Emirates
5 shared publications
Latest Publications
- Open Access
- 8 October 2026
- Published by Elsevier in SSRN Electronic Journal
Predicting Sustainable Consumption Behavior under Incomplete Information: Comparing Machine Learning and a Fine-Tuned Large Language Model
Predicting human behavior from structured survey data is challenging because behavioral outcomes may depend on nonlinear interactions among psychological, social, product-related, and contextual factors, while respondent information may also be incomplete. This study compares conventional tabular machine learning with a fine-tuned large language model (LLM) for out-of-sample prediction of self-reported sustainable consumption behavior (SCB) under increasing predictor missingness. Data comprised 1,260 observations from six food- and clothing-related surveys spanning purchase, utilization, and disposal. Fourteen regression algorithms and a mean-prediction baseline were trained and evaluated using repeated cross-validation, validation data, and a held-out test set. ExtraTrees achieved the lowest repeated-cross-validation mean absolute error (MAE = 0.2922) and, on a held-out test set of 189 respondents, produced MAE = 0.3015, RMSE = 0.3932, and R2 = 0.6873. In parallel, Qwen2.5-3B-Instruct, fine-tuned using QLoRA on controlled representations of respondent profiles, profiles, achieved MAE = 0.3319, RMSE = 0.4544, and R2 = 0.5545 at 0% imposed psychometric missingness and showed positive prediction bias. ExtraTrees remained more accurate up to 40% missingness. At 60% missingness, the LLM yielded lower MAE than ExtraTrees (0.3891 versus 0.4285), although RMSE and R2 were nearly identical between the two pipelines. The findings favor ExtraTrees for complete or moderately incomplete tabular data while identifying the missingness-trained LLM as a complementary predictor. The findings indicate that native tabular ensemble learning remains stronger under complete and moderately incomplete information, whereas missingness-trained LLMs may offer complementary value under severe information loss. The findings show that predictive utility depends on both information completeness and how available information is represented.
- 8 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPredicting Sustainable Consumption Behavior under Incomplete Information: Comparing Machine Learning and a Fine-Tuned Large Language Model
Predicting human behavior from structured survey data is challenging because behavioral outcomes may depend on nonlinear interactions among psychological, social, product-related, and contextual factors, while respondent information may also be incomplete. This study compares conventional tabular machine learning with a fine-tuned large language model (LLM) for out-of-sample prediction of self-reported sustainable consumption behavior (SCB) under increasing predictor missingness. Data comprised 1,260 observations from six food- and clothing-related surveys spanning purchase, utilization, and disposal. Fourteen regression algorithms and a mean-prediction baseline were trained and evaluated using repeated cross-validation, validation data, and a held-out test set. ExtraTrees achieved the lowest repeated-cross-validation mean absolute error (MAE = 0.2922) and, on a held-out test set of 189 respondents, produced MAE = 0.3015, RMSE = 0.3932, and R2 = 0.6873. In parallel, Qwen2.5-3B-Instruct, fine-tuned using QLoRA on controlled representations of respondent profiles, profiles, achieved MAE = 0.3319, RMSE = 0.4544, and R2 = 0.5545 at 0% imposed psychometric missingness and showed positive prediction bias. ExtraTrees remained more accurate up to 40% missingness. At 60% missingness, the LLM yielded lower MAE than ExtraTrees (0.3891 versus 0.4285), although RMSE and R2 were nearly identical between the two pipelines. The findings favor ExtraTrees for complete or moderately incomplete tabular data while identifying the missingness-trained LLM as a complementary predictor. The findings indicate that native tabular ensemble learning remains stronger under complete and moderately incomplete information, whereas missingness-trained LLMs may offer complementary value under severe information loss. The findings show that predictive utility depends on both information completeness and how available information is represented.
- Open Access
MMH: A multimodal dataset of whole-body kinematics, bilateral ground reaction forces, and lower-limb surface electromyography signals during load lifting and lowering
This study presents the MMH dataset, a laboratory-collected in vivo dataset comprising whole-body kinematics, three-dimensional ground reaction forces and two-dimensional centres of pressure under both feet, as well as surface electromyography (sEMG) signals of twelve lower-limb muscles (six muscles per leg) during load lifting and lowering tasks. Ten healthy, normal-weight, young male adults each performed 72 trials combining one- and two-handed load (2 kg) lifting and lowering. These trials include multiple initial and final load locations while using three different lifting techniques (stoop, semi-squat, and full-squat). The kinematic and force-plate measurements provide rich input for ergonomic risk assessment tools and optimisation-based musculoskeletal models aimed at quantifying and managing musculoskeletal risk of injury. Also, the sEMG recordings enable the development of EMG-assisted musculoskeletal models and support validation of predictions from optimisation-based models. These makes the multimodal MMH dataset a valuable resource for biomechanics, ergonomics, and human movement research.
Open AccessMMH: A multimodal dataset of whole-body kinematics, bilateral ground reaction forces, and lower-limb surface electromyography signals during load lifting and lowering
This study presents the MMH dataset, a laboratory-collected in vivo dataset comprising whole-body kinematics, three-dimensional ground reaction forces and two-dimensional centres of pressure under both feet, as well as surface electromyography (sEMG) signals of twelve lower-limb muscles (six muscles per leg) during load lifting and lowering tasks. Ten healthy, normal-weight, young male adults each performed 72 trials combining one- and two-handed load (2 kg) lifting and lowering. These trials include multiple initial and final load locations while using three different lifting techniques (stoop, semi-squat, and full-squat). The kinematic and force-plate measurements provide rich input for ergonomic risk assessment tools and optimisation-based musculoskeletal models aimed at quantifying and managing musculoskeletal risk of injury. Also, the sEMG recordings enable the development of EMG-assisted musculoskeletal models and support validation of predictions from optimisation-based models. These makes the multimodal MMH dataset a valuable resource for biomechanics, ergonomics, and human movement research.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 34, 4384-4394
Beyond BMI: The Role of Anthropometric Measurements in Predicting Plantar Pressure Distribution and Gait Biomechanics
Plantar pressure (PP) is a key biomechanical parameter for assessing foot function, gait abnormalities, and injury risk. Although body mass index (BMI) correlates with PP magnitude/distribution during gait, it is a coarse metric that collapses body composition into a ratio of mass to height, implicitly assuming uniform body proportions and mass distribution. The effects of individualized anthropometric parameters on PP signatures remain under-characterized, with a gap in incorporating body-shape measures beyond BMI. This exploratory study examined associations among anthropometric measurements, plantar peak pressure (PPP) outcomes, and model-predicted lower-limb joint kinetics. Data from 21 participants (normal-weight, overweight, obese) were collected using Xsens inertial measurement units and a Tekscan walkway. Joint kinetics were predicted with the AnyBody musculoskeletal model using two upper-body mass scaling approaches: constant-percentage-based (CPB) and body-shape-based (BSB). Pearson correlation analysis showed BMI had limited association with plantar peak pressure (PPP), yielding only two significant correlations across groups. In contrast, anthropometric measures, particularly limb lengths and circumferences, showed more frequent associations. Lower arm and shoulder lengths each exhibited up to five significant correlations, while chest, thigh, ankle, and wrist circumferences were meaningful. Compared with CPB, BSB yielded more frequent and generally larger-magnitude PPP-kinetic correlations, indicating that anthropometry-informed scaling altered the observed association patterns. In the normal-weight group, BSB identified seven significant correlations versus two with CPB; in the obese group, correlations exceeding 0.9 occurred nine times with BSB versus three with CPB. These results highlight the limited and inconsistent associations of BMI with PPP and support further investigation of detailed anthropometry in individualized PP and gait assessment.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Vol. 34, 4384-4394
Open AccessBeyond BMI: The Role of Anthropometric Measurements in Predicting Plantar Pressure Distribution and Gait Biomechanics
Plantar pressure (PP) is a key biomechanical parameter for assessing foot function, gait abnormalities, and injury risk. Although body mass index (BMI) correlates with PP magnitude/distribution during gait, it is a coarse metric that collapses body composition into a ratio of mass to height, implicitly assuming uniform body proportions and mass distribution. The effects of individualized anthropometric parameters on PP signatures remain under-characterized, with a gap in incorporating body-shape measures beyond BMI. This exploratory study examined associations among anthropometric measurements, plantar peak pressure (PPP) outcomes, and model-predicted lower-limb joint kinetics. Data from 21 participants (normal-weight, overweight, obese) were collected using Xsens inertial measurement units and a Tekscan walkway. Joint kinetics were predicted with the AnyBody musculoskeletal model using two upper-body mass scaling approaches: constant-percentage-based (CPB) and body-shape-based (BSB). Pearson correlation analysis showed BMI had limited association with plantar peak pressure (PPP), yielding only two significant correlations across groups. In contrast, anthropometric measures, particularly limb lengths and circumferences, showed more frequent associations. Lower arm and shoulder lengths each exhibited up to five significant correlations, while chest, thigh, ankle, and wrist circumferences were meaningful. Compared with CPB, BSB yielded more frequent and generally larger-magnitude PPP-kinetic correlations, indicating that anthropometry-informed scaling altered the observed association patterns. In the normal-weight group, BSB identified seven significant correlations versus two with CPB; in the obese group, correlations exceeding 0.9 occurred nine times with BSB versus three with CPB. These results highlight the limited and inconsistent associations of BMI with PPP and support further investigation of detailed anthropometry in individualized PP and gait assessment.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 133947-133969
Sex-Specific Pelvis Geometry Influences Musculoskeletal Gait Simulations: Quantifying Bias in Male-Based Models
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 133947-133969
Open AccessSex-Specific Pelvis Geometry Influences Musculoskeletal Gait Simulations: Quantifying Bias in Male-Based Models
- Open Access
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 237-244
Multimodal Assessment of Orthotic Intervention in Flatfoot Gait Using EMG and Plantar Pressure Analysis
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 237-244
Open AccessMultimodal Assessment of Orthotic Intervention in Flatfoot Gait Using EMG and Plantar Pressure Analysis
- Open Access
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 218-224
Prediction of Knee Contact Force During Gait using Artificial Intelligence: A Systematic Review and Meta-Analysis
Accurate estimation of knee contact forces (KCFs) is essential for planning treatments and designing implants for musculoskeletal (MSK) conditions such as osteoarthritis. Traditional methods, including instrumented implants and MSK simulations, are costly, complex, and time-consuming. Recent advances in artificial intelligence (AI) offer a more efficient alternative, enabling KCF prediction using kinematics, ground reaction forces, or muscle activity. This study aimed at assessing the effectiveness of AI-based models in predicting total and medial KCFs during normal and altered gait conditions (medial thrust, trunk sway, and walking with poles), compared to reference values from implants or MSK modeling. A systematic review and meta-analysis of ten studies was conducted following PRISMA guidelines and using a random-effects model. Study quality and bias were evaluated using the PROBAST+AI tool. Results indicated strong AI model performance for medial KCF prediction under both normal and altered gait, with pooled correlation coefficients above 0.9 and RMSE values under 30% BW. However, predictions for total KCF during normal gait showed higher error, highlighting challenges in model generalizability.
- 27 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 218-224
Open AccessPrediction of Knee Contact Force During Gait using Artificial Intelligence: A Systematic Review and Meta-Analysis
Accurate estimation of knee contact forces (KCFs) is essential for planning treatments and designing implants for musculoskeletal (MSK) conditions such as osteoarthritis. Traditional methods, including instrumented implants and MSK simulations, are costly, complex, and time-consuming. Recent advances in artificial intelligence (AI) offer a more efficient alternative, enabling KCF prediction using kinematics, ground reaction forces, or muscle activity. This study aimed at assessing the effectiveness of AI-based models in predicting total and medial KCFs during normal and altered gait conditions (medial thrust, trunk sway, and walking with poles), compared to reference values from implants or MSK modeling. A systematic review and meta-analysis of ten studies was conducted following PRISMA guidelines and using a random-effects model. Study quality and bias were evaluated using the PROBAST+AI tool. Results indicated strong AI model performance for medial KCF prediction under both normal and altered gait, with pooled correlation coefficients above 0.9 and RMSE values under 30% BW. However, predictions for total KCF during normal gait showed higher error, highlighting challenges in model generalizability.
- 27 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 145-149
Sagittal Reaction Moments During Gait using Kinematic Inputs: A Deep Learning Approach
Accurate estimation of joint reaction moments (JRMs) and ground reaction moments (GRMs) during gait is critical for diagnosing musculoskeletal disorders but remains constrained by lab-dependent methods requiring force plates or full-body kinematics. This study introduces a novel deep learning framework to predict sagittal plane JRMs and GRMs using only single-marker positional data (hip, knee, ankle), eliminating the need for complex instrumentation. Leveraging sequence-to-sequence models (GRU, LSTM, 1D-CNN) optimized via Bayesian hyperparameter tuning and Leave-One-Subject-Out (LOSO) cross-validation, we analyzed gait data from 80 asymptomatic adults. Models achieved RMSE values of $31.1-50.1 \text{Nmm} / \text{kg}$ across joints, with normalized RMSE (nRMSE) between $\mathbf{1 6} \boldsymbol{-} \mathbf{2 1 {\%}}$, comparable to prior studies using more extensive inputs. GRU excelled at ankle moment prediction $(31.1 \pm 25.8$ vs. $38.6 \pm 30.0$ Nmm/kg, left/right), while LSTM showed robustness for knee moments $(35.0 \pm 25.6$ vs. $37.2 \pm 30.1 \text{Nmm} / \text{kg})$. Asymmetries in hip RMSE $(48.7 \pm 25.7 \mathrm{L}$ vs. $42.4 \pm 23.4 \mathrm{R})$ highlighted joint-specific variability. The framework's minimal kinematic requirements reduce computational burden, offering a scalable alternative to traditional gait analysis. Despite strong performance in asymptomatic individuals, validation in pathological populations and integration with Markerless systems are needed for clinical translation. This work advances accessible, lab-free biomechanical assessment, demonstrating the potential of deep learning to bridge gaps between research and clinical practice.- 27 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 145-149
Sagittal Reaction Moments During Gait using Kinematic Inputs: A Deep Learning Approach
Accurate estimation of joint reaction moments (JRMs) and ground reaction moments (GRMs) during gait is critical for diagnosing musculoskeletal disorders but remains constrained by lab-dependent methods requiring force plates or full-body kinematics. This study introduces a novel deep learning framework to predict sagittal plane JRMs and GRMs using only single-marker positional data (hip, knee, ankle), eliminating the need for complex instrumentation. Leveraging sequence-to-sequence models (GRU, LSTM, 1D-CNN) optimized via Bayesian hyperparameter tuning and Leave-One-Subject-Out (LOSO) cross-validation, we analyzed gait data from 80 asymptomatic adults. Models achieved RMSE values of $31.1-50.1 \text{Nmm} / \text{kg}$ across joints, with normalized RMSE (nRMSE) between $\mathbf{1 6} \boldsymbol{-} \mathbf{2 1 {\%}}$, comparable to prior studies using more extensive inputs. GRU excelled at ankle moment prediction $(31.1 \pm 25.8$ vs. $38.6 \pm 30.0$ Nmm/kg, left/right), while LSTM showed robustness for knee moments $(35.0 \pm 25.6$ vs. $37.2 \pm 30.1 \text{Nmm} / \text{kg})$. Asymmetries in hip RMSE $(48.7 \pm 25.7 \mathrm{L}$ vs. $42.4 \pm 23.4 \mathrm{R})$ highlighted joint-specific variability. The framework's minimal kinematic requirements reduce computational burden, offering a scalable alternative to traditional gait analysis. Despite strong performance in asymptomatic individuals, validation in pathological populations and integration with Markerless systems are needed for clinical translation. This work advances accessible, lab-free biomechanical assessment, demonstrating the potential of deep learning to bridge gaps between research and clinical practice.- Open Access
- 27 January 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 186, 109739
Musculoskeletal model predictions sensitivity to upper body mass scaling during gait
- 27 January 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 186, 109739
Open AccessMusculoskeletal model predictions sensitivity to upper body mass scaling during gait
- Open Access
- 13 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 188-192
Seq2Seq Deep Learning Models to Predict Ground Reaction Forces During Gait
- 13 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 188-192
Open AccessSeq2Seq Deep Learning Models to Predict Ground Reaction Forces During Gait
- Open Access
- 13 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 173-177
Assessing Gait Adaptation to Prehabilitation in ACL-Injured Patients Using Kinematic and Kinetic Analyses
- 13 November 2024
- Published by Association for Computing Machinery (ACM)
- p. 173-177
Open AccessAssessing Gait Adaptation to Prehabilitation in ACL-Injured Patients Using Kinematic and Kinetic Analyses