Abdellah Madani
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Abdellah Madani has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Abdelali Elkouay
University of Chouaib Doukkali, Faculty of Science EL Jadida
1 shared publication - Chaimaa Bouafoud
LaROSERI Laboratory, Chouaib Doukkali University, El Jadida, Morocco
1 shared publication - Douha Tissir
Department of Computer Science, Chouaib Doukkali University, El Jadida, Morocco, Morocco
1 shared publication - Fatima Ezzahra El Kamouny
Department of Computer Science,Chouaib Doukkali University,El Jadida,Morocco, Morocco
1 shared publication - Khadija El Kamouny
Green Tech Institute Department, Université Mohammed VI Polytechnique, Marrakesh, Morocco, Morocco
1 shared publication - Khalid Zine-Dine
Faculty of Sciences Rabat (FSR), Mohammed V University, Rabat, Morocco
1 shared publication
Latest Publications
- Open Access
- 12 October 2026
- Published by MDPI AG in Preprints.org
Prediction of Creatinine from NIR Spectroscopy Using Chemometric and Deep Learning Models with Indirect Urea Estimation
Renal biomarker estimation is important for evaluating kidney health status. Traditional methods for determining creatinine and urea levels rely on chemical reagents and analytical instruments. This study explores an alternative, reagent-free approach for determining creatinine concentration and estimating urea levels in human plasma samples using near-infrared (NIR) spectroscopy, chemometrics, and artificial intelligence. NIR spectra were acquired from 500 blood samples using a portable NIR spectrometer operating between 1000 and 2600 nm, with laboratory-determined creatinine and urea concentrations used as reference values for calibration. The spectra were preprocessed using Multiplicative Scatter Correction (MSC), Standard Normal Variate (SNV), and Savitzky–Golay (SG) filtering, which improved the PLS coefficient of determination from R2=0.72 to R2=0.94. Interval Partial Least Squares (iPLS) was then used to select the most informative wavelengths, yielding five spectral intervals: 1020–1060, 1430–1480, 1590–1650, 1950–2020, and 2080–2150 nm. Creatinine concentration was predicted using Partial Least Squares (PLS), a Generative Adversarial Network (GAN), a Convolutional Neural Network (CNN), and a Long Short-Term Memory (LSTM) network, with the data split into 400 training and 100 test observations. On the held-out internal test set, the GAN achieved the best performance among all tested methods, with R2=0.97, RMSE of 1.31 mg/L, and MAE of 0.94 mg/L, while the CNN, LSTM, and PLS models achieved R2 values of 0.95, 0.94, and 0.93, respectively. An indirect method for urea estimation, based on the statistical correlation between creatinine and urea, was also examined: the NIR-predicted creatinine levels were used in a linear regression model relating creatinine and urea, yielding R2=0.92, RMSE = 0.18 g/L, and MAE = 0.13 g/L. Overall, these findings demonstrate the feasibility of combining NIR spectroscopy, spectral preprocessing, iPLS-based wavelength selection, and deep learning for the analysis of renal biomarkers. The proposed NIR-iPLS-GAN framework successfully predicted creatinine concentration and enabled the extrapolation of spectral information to indirectly estimate urea concentration.
- 12 October 2026
- Published by MDPI AG in Preprints.org
Open AccessPrediction of Creatinine from NIR Spectroscopy Using Chemometric and Deep Learning Models with Indirect Urea Estimation
Renal biomarker estimation is important for evaluating kidney health status. Traditional methods for determining creatinine and urea levels rely on chemical reagents and analytical instruments. This study explores an alternative, reagent-free approach for determining creatinine concentration and estimating urea levels in human plasma samples using near-infrared (NIR) spectroscopy, chemometrics, and artificial intelligence. NIR spectra were acquired from 500 blood samples using a portable NIR spectrometer operating between 1000 and 2600 nm, with laboratory-determined creatinine and urea concentrations used as reference values for calibration. The spectra were preprocessed using Multiplicative Scatter Correction (MSC), Standard Normal Variate (SNV), and Savitzky–Golay (SG) filtering, which improved the PLS coefficient of determination from R2=0.72 to R2=0.94. Interval Partial Least Squares (iPLS) was then used to select the most informative wavelengths, yielding five spectral intervals: 1020–1060, 1430–1480, 1590–1650, 1950–2020, and 2080–2150 nm. Creatinine concentration was predicted using Partial Least Squares (PLS), a Generative Adversarial Network (GAN), a Convolutional Neural Network (CNN), and a Long Short-Term Memory (LSTM) network, with the data split into 400 training and 100 test observations. On the held-out internal test set, the GAN achieved the best performance among all tested methods, with R2=0.97, RMSE of 1.31 mg/L, and MAE of 0.94 mg/L, while the CNN, LSTM, and PLS models achieved R2 values of 0.95, 0.94, and 0.93, respectively. An indirect method for urea estimation, based on the statistical correlation between creatinine and urea, was also examined: the NIR-predicted creatinine levels were used in a linear regression model relating creatinine and urea, yielding R2=0.92, RMSE = 0.18 g/L, and MAE = 0.13 g/L. Overall, these findings demonstrate the feasibility of combining NIR spectroscopy, spectral preprocessing, iPLS-based wavelength selection, and deep learning for the analysis of renal biomarkers. The proposed NIR-iPLS-GAN framework successfully predicted creatinine concentration and enabled the extrapolation of spectral information to indirectly estimate urea concentration.
- Open Access
- 11 September 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 20 (17) , 4–25
Developing a Chatbot for Soft Skills Education: Comparing ReLU-LSTM and GloAT-Transformer Models
Recently, soft skills, like communication, teamwork, and problem-solving, have become more important in higher education. In Moroccan universities, fostering these skills is essential to better prepare students for the demands of the job market and modern work environments. However, traditional educational methods often lack interactive and personalized approaches to develop these competencies. This led us to develop a conversational AI system to support soft skills learning through intent classification using Natural Language Processing (NLP) and Neural Networks. Two deep learning models were developed and evaluated; a ReLU-Enhanced LSTM, optimized for stable sequential processing, and a GloAT-Transformer model leveraging global self-attention to capture contextual relationships.
- 11 September 2026
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 20 (17) , 4–25
Open AccessDeveloping a Chatbot for Soft Skills Education: Comparing ReLU-LSTM and GloAT-Transformer Models
Recently, soft skills, like communication, teamwork, and problem-solving, have become more important in higher education. In Moroccan universities, fostering these skills is essential to better prepare students for the demands of the job market and modern work environments. However, traditional educational methods often lack interactive and personalized approaches to develop these competencies. This led us to develop a conversational AI system to support soft skills learning through intent classification using Natural Language Processing (NLP) and Neural Networks. Two deep learning models were developed and evaluated; a ReLU-Enhanced LSTM, optimized for stable sequential processing, and a GloAT-Transformer model leveraging global self-attention to capture contextual relationships.
- 3 July 2022
- Published by Springer Nature
- p. 85-99
Classification of URLs Using N-gram Machine Learning Approach
- 3 July 2022
- Published by Springer Nature
- p. 85-99
Classification of URLs Using N-gram Machine Learning Approach