Lamiae Eloutouate

Dr. Lamiae Eloutouate

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
Verified
Journal Article
Open access
PDF file
TBTajeddine BENBARRAD
Tajeddine BENBARRAD
LELamiae Eloutouate
Lamiae Eloutouate
MAMounir Arioua
Mounir Arioua
FEFatiha Elouaai
Fatiha Elouaai
MyDriss LAANAOUIMyDriss LAANAOUI
MyDriss LAANAOUI
Machine vision is increasingly replacing manual steel surface inspection. The automatic inspection of steel surface defects makes it possible to ensure the quality of products in the steel industry with high accuracy. However, the optimization of inspection time presents a great challenge for the integration of machine vision in high-speed production lines. In this context, compressing the collected images before transmission is essential to save bandwidth and energy, and improve the latency of vision applications. The aim of this paper was to study the impact of quality degradation resulting from image compression on the classification performance of steel surface defects with a CNN. Image compression was applied to the Northeastern University (NEU) surface-defect database with various compression ratios. Three different models were trained and tested with these images to classify surface defects using three different approaches. The obtained results showed that trained and tested models on the same compression qualities maintained approximately the same classification performance for all used compression grades. In addition, the findings clearly indicated that the classification efficiency was affected when the training and test datasets were compressed using different parameters. This impact was more obvious when there was a large difference between these compression parameters, and for models that achieved very high accuracy. Finally, it was found that compression-based data augmentation significantly increased the classification precision to perfect scores (98–100%), and thus improved the generalization of models when tested on different compression qualities. The importance of this work lies in exploiting the obtained results to successfully integrate image compression into machine vision systems, and as appropriately as possible.

1 August 2023 in International Journal of Online and Biomedical Engineering (iJOE)
Verified
Journal Article
Open access
PDF file
Hicham GIBET TANIHicham GIBET TANI
Hicham GIBET TANI
LELamiae Eloutouate
Lamiae Eloutouate
FEFatiha Elouaai
Fatiha Elouaai
MBMohammed Bouhorma
Mohammed Bouhorma
MHMohamed Walid Hajoub
Mohamed Walid Hajoub
Image captioning is a promising technique for remote monitoring of patient behavior, enabling healthcare providers to identify changes in patient routines and conditions. In this study, we explore the use of transformer neural networks for image caption generation from surveillance camera footage, captured at regular intervals of one minute. Our goal is to develop and evaluate a transformer neural network model, trained and tested on the COCO (common objects in context) dataset, for generating captions that describe patient behavior. Furthermore, we will compare our proposed approach with a traditional convolutional neural network (CNN) method to highlight the prominence of our proposed approach. Our findings demonstrate the potential of transformer neural networks in generating natural language descriptions of patient behavior, which can provide valuable insights for healthcare providers. The use of such technology can allow for more efficient monitoring of patients, enabling timely interventions when necessary. Moreover, our study highlights the potential of transformer neural networks in identifying patterns and trends in patient behavior over time, which can aid in developing personalized healthcare plans.

24 May 2023
Verified
Conference Paper
Open access
LELamiae Eloutouate
Lamiae Eloutouate
Hicham GIBET TANIHicham GIBET TANI
Hicham GIBET TANI
EFElouaai Fatiha
Elouaai Fatiha
MBMohammed Bouhorma
Mohammed Bouhorma
MHMohamed Walid Hajoub
Mohamed Walid Hajoub
Image captioning is a task that involves generating natural language descriptions of the content of an image, and has the potential to support healthcare providers in monitoring patient conditions and routines at home. The ability to remotely monitor patients can provide valuable information to healthcare providers, allowing them to identify changes in patient behavior and facilitate timely interventions. In this study, we examine the usability of transformer neural networks for image caption generation from surveillance camera footage taken at regular intervals of one minute. Our objective is to develop and evaluate a transformer neural network model for generating captions of patient behavior, trained and evaluated on the Common Objects in Context (COCO) dataset. Our study provides a proof-of-concept for the potential of transformer neural networks in image captioning for remote monitoring of patient behavior. By generating natural language descriptions of patient behavior, healthcare providers can obtain valuable insights into patient routines and conditions, allowing them to monitor patients remotely and identify changes in behavior that may require intervention. Furthermore, our study highlights the potential for transformer neural networks to support healthcare providers in identifying patterns and trends in patient behavior over time.

25 August 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
LELamiae Eloutouate
Lamiae Eloutouate
Hicham GIBET TANIHicham GIBET TANI
Hicham GIBET TANI
Background and Objective. Cardiovascular disease (CVD) prediction studies based on the UCI Heart Disease Dataset frequently rely on evaluation protocols that do not fully account for source-level data leakage and often emphasize discrimination metrics while overlooking fairness, calibration, and explainability. This work quantifies the impact of leakage prevention and examines the behaviour of six machine learning classifiers across multiple evaluation dimensions using independent external cohorts.Methods. Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), XGBoost, CatBoost, and LightGBM were trained on UCI-derived data (n = 916) after excluding Cleveland (n = 303) and Statlog (n = 267) records prior to any preprocessing. External evaluation was conducted using the Cleveland Heart Disease Registry, Statlog Heart Disease Dataset, and Framingham Heart Study Teaching Dataset (n = 4,240). Explainability was assessed using SHAP TreeExplainer for ensemble models, and attribution consistency was examined against established cardiovascular risk factors. A benchmark-specific composite index, the Clinical Explainability Trust Score (CETS), was used to summarize discrimination, attribution stability, guideline concordance, and sex-based fairness.Results. Under leakage-free conditions, no classifier achieved an internal AUC of 0.80, contrasting with the 0.92–0.96 range frequently reported in leakage-affected benchmarks. SHAP analyses consistently highlighted the same five predictors across all ensemble models, indicating complete cross-model agreement in feature ranking. Ensemble classifiers exhibited substantially lower Equal Opportunity Difference (EOD < 0.25) than LR and SVM (EOD > 0.78). LightGBM obtained the highest composite score (0.505), whereas LR achieved the highest mean external AUC (0.693) but ranked lowest on the composite assessment because of elevated fairness disparity (EOD = 0.784).Conclusions. Models with similar levels of external discrimination displayed notable differences in fairness, calibration, and attribution stability. LightGBM and CatBoost produced the most balanced overall evaluation profiles and represent promising candidates for future prospective clinical investigation. These findings suggest that reliance on external AUC alone may provide an incomplete view of model behaviour in cardiovascular risk prediction and support the use of multidimensional evaluation strategies in future studies.

24 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
LELamiae Eloutouate
Lamiae Eloutouate
Hicham GIBET TANIHicham GIBET TANI
Hicham GIBET TANI
Cardiovascular disease (CVD) prediction studies based on the UCI Heart Disease Dataset frequently rely on evaluation protocols that do not fully account for source-level data leakage and often emphasize discrimination metrics while overlooking fairness, calibration,and explainability. This work quantifies the impact of leakage prevention and examines the behaviour of six machine learning classifiers across multiple evaluation dimensions using independent external cohorts.Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), XGBoost, CatBoost, and LightGBM were trained on UCI-derived data (n = 916) after excluding Cleveland (n = 303) and Statlog (n = 267) records prior to any preprocessing. External evaluation was conducted using the Cleveland Heart Disease Registry, Statlog Heart Disease Dataset, and Framingham Heart Study Teaching Dataset (n = 4,240). Explainability was assessed using SHAP TreeExplainer for ensemble models, and attribution consistency was examined against established cardiovascular risk factors. A benchmark-specific composite index, the Clinical Explainability Trust Score (CETS), was used to summarize discrimination, attribution stability, guideline concordance, and sex-based fairness.Under leakage-free conditions, no classifier achieved an internal AUC of 0.80, contrasting with the 0.92–0.96 range frequently reported in leakage-affected benchmarks. SHAP analyses consistently highlighted the same five predictors across all ensemble models, indicating complete cross-model agreement in feature ranking. Ensemble classifiers exhibited substantially lower Equal Opportunity Difference (EOD < 0.25) than LR and SVM (EOD > 0.78). LightGBM obtained the highest composite score (0.505), whereas LR achieved the highest mean external AUC (0.693) but ranked lowest on the composite assessment because of elevated fairness disparity (EOD = 0.784).

Verified
Conference Paper
Open access
PDF file
LELamiae Eloutouate
Lamiae Eloutouate
Hicham GIBET TANIHicham GIBET TANI
Hicham GIBET TANI
LELotfi Elaachak
Lotfi Elaachak
FEFatiha Elouaai
Fatiha Elouaai
MBMohammed Bouhorma
Mohammed Bouhorma
The application of machine learning (ML) to medical datasets offers significant potential for improving disease prediction and patient outcomes. However, challenges such as feature redundancy, overfitting, and suboptimal model performance limit the practical effectiveness of ML algorithms. This study focuses on optimizing ML techniques for cardiovascular disease prediction using the Kaggle Cardiovascular Disease dataset. We systematically apply feature selection methods, including correlation analysis and regularization techniques (L1/L2), to identify the most relevant attributes and address multicollinearity. Advanced ensemble models such as Random Forest, XGBoost, and LightGBM are employed to mitigate overfitting and enhance predictive performance. Through hyperparameter tuning and stratified k-fold cross-validation, we ensure model robustness and generalizability. The results demonstrate that ensemble methods, particularly gradient boosting algorithms, outperform traditional models, achieving superior predictive accuracy and stability. This study highlights the importance of algorithm optimization in ML applications for healthcare, offering a replicable framework for medical datasets and paving the way for more effective diagnostic tools in cardiovascular health.