Soumia Ziti
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Soumia Ziti has a total of 185 co-authors. Here are the top co-authors by number of shared publications:
- Lahoucine Bahmad
Laboratory of Condensed Matter and Interdisciplinary Sciences (LaMCScI), Faculty of Sciences, Mohammed V University, Av. Ibn Batouta, B. P. 1014, Rabat, Morocco
21 shared publications - Nassim Kharmoum
Mohammed V University in Rabat
17 shared publications H. Labrim
15 shared publications- Yassine Zaoui Seghroucheni
Mohammed V University in Rabat
9 shared publications - Abdeslam El Harraj
Moroccan National Society for Television and Radio Broadcasting (SNRT)
7 shared publications - Btissam Acim
Mohammed V University in Rabat
7 shared publications
Latest Publications
- Open Access
- 13 November 2026
- Published by MDPI AG in Information
- Vol. 17 (11) , 1145
An Interpretable Machine Learning Comparison for Employee Performance Prediction: Implications for Decision Support in Corporate Learning
Identifying employees who may need additional professional-learning support could help organisations allocate training resources, but model comparisons in this area often omit trivial baselines and class-specific metrics. We compare seven classifiers (decision tree, random forest, logistic regression, k-nearest neighbours, XGBoost, LightGBM, CatBoost) with a rule-based and a majority-class baseline under one protocol: stratified hold-out, the same tuning procedure, paired bootstrap confidence intervals, a descriptive repeated cross-validation check, and SHAP and LIME explanations. Experiment 1 predicts a KPI label of undocumented origin for 17,415 employees (35.9% positive); it is an illustrative comparison using the label as distributed, not validated evidence about KPI attainment. Experiment 2 predicts promotion for 54,808 employees (8.5% positive), a related task from the same public source and not an external validation. In Experiment 1, learning models reached macro-F1 of 0.628 to 0.663, against 0.590 for the rule-based baseline; a depth-5 decision tree was 0.005 below the reference random forest (95% bootstrap CI −0.016 to +0.005) but was not shown non-inferior at a 0.01 margin. In Experiment 2, boosting models were best (macro-F1 0.734 to 0.739), the decision tree was clearly worse (−0.030), and only about 35% of promoted employees were identified. Associations with training indicators are not causal, and no effect on decisions was tested.
- 13 November 2026
- Published by MDPI AG in Information
- Vol. 17 (11) , 1145
Open AccessAn Interpretable Machine Learning Comparison for Employee Performance Prediction: Implications for Decision Support in Corporate Learning
Identifying employees who may need additional professional-learning support could help organisations allocate training resources, but model comparisons in this area often omit trivial baselines and class-specific metrics. We compare seven classifiers (decision tree, random forest, logistic regression, k-nearest neighbours, XGBoost, LightGBM, CatBoost) with a rule-based and a majority-class baseline under one protocol: stratified hold-out, the same tuning procedure, paired bootstrap confidence intervals, a descriptive repeated cross-validation check, and SHAP and LIME explanations. Experiment 1 predicts a KPI label of undocumented origin for 17,415 employees (35.9% positive); it is an illustrative comparison using the label as distributed, not validated evidence about KPI attainment. Experiment 2 predicts promotion for 54,808 employees (8.5% positive), a related task from the same public source and not an external validation. In Experiment 1, learning models reached macro-F1 of 0.628 to 0.663, against 0.590 for the rule-based baseline; a depth-5 decision tree was 0.005 below the reference random forest (95% bootstrap CI −0.016 to +0.005) but was not shown non-inferior at a 0.01 margin. In Experiment 2, boosting models were best (macro-F1 0.734 to 0.739), the decision tree was clearly worse (−0.030), and only about 35% of promoted employees were identified. Associations with training indicators are not causal, and no effect on decisions was tested.
- Open Access
- 12 November 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 168 (1) , 1
Representation over Resampling: A Causal Lightweight Approach for IMU-Only UAV Fault Detection
Predictive maintenance for Unmanned Aerial Vehicles (UAVs) faces severe class imbalance challenges. While algorithmic resampling (SMOTE, ADASYN) is commonly used to mitigate this, we demonstrate that for high-frequency Inertial Measurement Unit (IMU) telemetry, such methods are ineffective and actively degrade performance by generating physically impossible “chimeric” fault trajectories. Through a systematic benchmark of 1161 flights, we demonstrate that representation, not class balance, is the primary bottleneck, and propose Causal Lightweight Feature Engineering (CL-FE), extracting 18 temporal descriptors over a 1.0 s causal window. CL-FE drastically outperforms class-balancing efforts, delivering a +36% relative increase in PR-AUC over raw data and improving imminent fault recall by 11 percentage points at safety-critical false alarm rates (FPR ≤ 2%). SHAP analysis confirms these descriptors capture the actual physical vibration signatures of incipient degradation. With a negligible 0.72 ms latency, CL-FE proves that physics-informed representation is vastly superior to algorithmic balancing for embedded real-time UAV deployment.
- 12 November 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 168 (1) , 1
Open AccessRepresentation over Resampling: A Causal Lightweight Approach for IMU-Only UAV Fault Detection
Predictive maintenance for Unmanned Aerial Vehicles (UAVs) faces severe class imbalance challenges. While algorithmic resampling (SMOTE, ADASYN) is commonly used to mitigate this, we demonstrate that for high-frequency Inertial Measurement Unit (IMU) telemetry, such methods are ineffective and actively degrade performance by generating physically impossible “chimeric” fault trajectories. Through a systematic benchmark of 1161 flights, we demonstrate that representation, not class balance, is the primary bottleneck, and propose Causal Lightweight Feature Engineering (CL-FE), extracting 18 temporal descriptors over a 1.0 s causal window. CL-FE drastically outperforms class-balancing efforts, delivering a +36% relative increase in PR-AUC over raw data and improving imminent fault recall by 11 percentage points at safety-critical false alarm rates (FPR ≤ 2%). SHAP analysis confirms these descriptors capture the actual physical vibration signatures of incipient degradation. With a negligible 0.72 ms latency, CL-FE proves that physics-informed representation is vastly superior to algorithmic balancing for embedded real-time UAV deployment.
- Open Access
Cross-Layer BLE Device-Type Identification via Locality-Sensitive Hashing for IoT Security
Bluetooth Low Energy (BLE) devices are ubiquitous across consumer, industrial, and healthcare IoT ecosystems, yet their resource-constrained nature typically precludes conventional cryptographic authentication, leaving device identification as a security primitive for device management, access control, and threat detection in IoT environments. Current BLE fingerprinting techniques are restricted by rigid cryptographic hashes, machine learning models that require periodic retraining to cover new devices, or single-layer feature sets vulnerable to missing data. In response to these limitations, we introduce a Nilsimsa-based identification approach that performs BLE device type/manufacturer identification through merging link-layer features and advertised GATT service information before network access is granted to cope with missing or inaccessible data. We analyzed various combinations of features from different layers to assess how merging them impacts identification accuracy. The exhaustive analysis uses 10 test objects and 10 reference hashes. The large-scale evaluation uses 5000 test records and 10,000 reference hashes covering 104 device type–manufacturer pairs. Key findings highlight that cross-layer features are important to cope with missing features. The minimum subset size was found at with a subset-coverage criterion: at least 95% of all subsets of 10 features reach an accuracy of 90% or more. Our proposal not only facilitates the seamless enrollment of new devices through reference-hash registration without retraining but also enables confidence-based anomaly detection using its similarity score, providing an incrementally extensible approach for BLE security monitoring. Overall, our approach advances BLE fingerprinting by introducing a schema-agnostic, retraining-free, cross-layer identification model tailored for the dynamic environment of IoT deployments.Open AccessCross-Layer BLE Device-Type Identification via Locality-Sensitive Hashing for IoT Security
Bluetooth Low Energy (BLE) devices are ubiquitous across consumer, industrial, and healthcare IoT ecosystems, yet their resource-constrained nature typically precludes conventional cryptographic authentication, leaving device identification as a security primitive for device management, access control, and threat detection in IoT environments. Current BLE fingerprinting techniques are restricted by rigid cryptographic hashes, machine learning models that require periodic retraining to cover new devices, or single-layer feature sets vulnerable to missing data. In response to these limitations, we introduce a Nilsimsa-based identification approach that performs BLE device type/manufacturer identification through merging link-layer features and advertised GATT service information before network access is granted to cope with missing or inaccessible data. We analyzed various combinations of features from different layers to assess how merging them impacts identification accuracy. The exhaustive analysis uses 10 test objects and 10 reference hashes. The large-scale evaluation uses 5000 test records and 10,000 reference hashes covering 104 device type–manufacturer pairs. Key findings highlight that cross-layer features are important to cope with missing features. The minimum subset size was found at with a subset-coverage criterion: at least 95% of all subsets of 10 features reach an accuracy of 90% or more. Our proposal not only facilitates the seamless enrollment of new devices through reference-hash registration without retraining but also enables confidence-based anomaly detection using its similarity score, providing an incrementally extensible approach for BLE security monitoring. Overall, our approach advances BLE fingerprinting by introducing a schema-agnostic, retraining-free, cross-layer identification model tailored for the dynamic environment of IoT deployments. - 27 October 2026
- Published by Springer Nature in Network Modeling Analysis in Health Informatics and Bioinformatics
- Vol. 15 (1) , 227
Integration of discriminative and generative approaches in the computational chemistry and drug discovery process
- 27 October 2026
- Published by Springer Nature in Network Modeling Analysis in Health Informatics and Bioinformatics
- Vol. 15 (1) , 227
Integration of discriminative and generative approaches in the computational chemistry and drug discovery process
- Open Access
- 24 September 2026
- Published by Springer Nature in Discover Education
- Vol. 5 (1) , 994
Behavioral clustering for adaptive learning object recommendation in higher education using learning analytics for personalized student success
Higher education institutions increasingly rely on Learning Management Systems to deliver instruction, yet many platforms employ one-size-fits-all content delivery that fails to accommodate heterogeneity in student engagement, motivation, and performance. This study presents a behavioral clustering framework that derives learner profiles from behavioral and psychological proxies and maps them to a proposed adaptive learning object recommendation framework. Using a dataset of 14,003 anonymized student records with 16 attributes, including 12 behavioral, psychological, and environmental clustering inputs, we applied K-Means++ clustering (k = 6), validated by the Elbow Method and Silhouette Analysis, and benchmarked against four competing algorithms. Cluster distinctiveness was characterized through ANOVA on Final Grade (primary outcome: F(5, 13,997) = 9365.49, p < 0.001, η 2 = 0.770; Exam Score sensitivity analysis: F(5, 13,997) = 6822.77, p < 0.001, η 2 = 0.709), Tukey HSD post hoc tests, and chi-square analysis with Cramér's V. Six distinct learner profiles emerged, spanning from Highly Engaged Achievers to At-Risk Learners, with a mean exam score difference of 40.17 points between the highest and lowest-performing clusters. An affinity-scoring model combined with an FSLSM mapping matrix and diversity regularization (λ = 0.30) was proposed as a conceptual framework to translate validated profiles into cluster-specific learning object recommendations. The clustering achieved robust separation (Silhouette = 0.62), and the recommendation component provides an interpretable mapping that awaits empirical validation in live LMS environments. Because the clustering inputs are self-reported proxies rather than direct LMS interaction traces, validation with LMS data remains necessary to assess applicability to observed learning behavior.
- 24 September 2026
- Published by Springer Nature in Discover Education
- Vol. 5 (1) , 994
Open AccessBehavioral clustering for adaptive learning object recommendation in higher education using learning analytics for personalized student success
Higher education institutions increasingly rely on Learning Management Systems to deliver instruction, yet many platforms employ one-size-fits-all content delivery that fails to accommodate heterogeneity in student engagement, motivation, and performance. This study presents a behavioral clustering framework that derives learner profiles from behavioral and psychological proxies and maps them to a proposed adaptive learning object recommendation framework. Using a dataset of 14,003 anonymized student records with 16 attributes, including 12 behavioral, psychological, and environmental clustering inputs, we applied K-Means++ clustering (k = 6), validated by the Elbow Method and Silhouette Analysis, and benchmarked against four competing algorithms. Cluster distinctiveness was characterized through ANOVA on Final Grade (primary outcome: F(5, 13,997) = 9365.49, p < 0.001, η 2 = 0.770; Exam Score sensitivity analysis: F(5, 13,997) = 6822.77, p < 0.001, η 2 = 0.709), Tukey HSD post hoc tests, and chi-square analysis with Cramér's V. Six distinct learner profiles emerged, spanning from Highly Engaged Achievers to At-Risk Learners, with a mean exam score difference of 40.17 points between the highest and lowest-performing clusters. An affinity-scoring model combined with an FSLSM mapping matrix and diversity regularization (λ = 0.30) was proposed as a conceptual framework to translate validated profiles into cluster-specific learning object recommendations. The clustering achieved robust separation (Silhouette = 0.62), and the recommendation component provides an interpretable mapping that awaits empirical validation in live LMS environments. Because the clustering inputs are self-reported proxies rather than direct LMS interaction traces, validation with LMS data remains necessary to assess applicability to observed learning behavior.
- 12 September 2026
- Published by Elsevier in International Journal of Medical Informatics
- Vol. 222, 106716
FHIR as an interoperability enabler for clinical applications: a narrative review of implementations, impacts, and future challenges
- 12 September 2026
- Published by Elsevier in International Journal of Medical Informatics
- Vol. 222, 106716
FHIR as an interoperability enabler for clinical applications: a narrative review of implementations, impacts, and future challenges
- 2 September 2026
- Published by Springer Nature
- p. 63-73
A Meta-learning Framework for a Self-adapting, Data-Driven Digital Twin for Urban Traffic Control
- 2 September 2026
- Published by Springer Nature
- p. 63-73
A Meta-learning Framework for a Self-adapting, Data-Driven Digital Twin for Urban Traffic Control
- 2 September 2026
- Published by Springer Nature
- p. 223-230
Applications of Artificial Intelligence in Smart Health: From Clinical Practice to Digital Epidemic Management
- 2 September 2026
- Published by Springer Nature
- p. 223-230
Applications of Artificial Intelligence in Smart Health: From Clinical Practice to Digital Epidemic Management
- 2 August 2026
- Published by Springer Nature
- p. 345-356
AI Guardians: The New Frontier of Microservices Security
- 2 August 2026
- Published by Springer Nature
- p. 345-356
AI Guardians: The New Frontier of Microservices Security
- 22 July 2026
- Published by Springer Nature
- p. 546-555
DeepHealthFaces: AI-Avatar-Based Approach for Health Equity Compared with Traditional Digital Health Tools
- 22 July 2026
- Published by Springer Nature
- p. 546-555
DeepHealthFaces: AI-Avatar-Based Approach for Health Equity Compared with Traditional Digital Health Tools