T. Kujani
Research Subjects
Research Interests
T. Kujani has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Sathiya Priya Selvaraj
Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Saveetha University,Department of Computer Science Engineering,Chennai,India
4 shared publications - M. Amanullah
Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India
3 shared publications - R. Bhuvanya
Sri Ramachandra Institute of Higher Education and Research
2 shared publications - A. M. Mukilan
Sri Ramachandra Institute of Higher Education and Research
1 shared publication - Adhimoolam Kannan
Department of Mathematics, Vel Tech Multi Tech Dr. Rangarajan Dr. Sakunthala Engineering College, Chennai 600062, India
1 shared publication - Annamalai Meenakshi
Department of Mathematics, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Avadi, Chennai 600062, India
1 shared publication
Latest Publications
- 2 August 2026
- Published by Springer Nature
- p. 63-74
BioReFineNet: A Hybrid Dense U-Net ++ Architecture for High-Precision Blood Cell Segmentation from Microscopic Slides
- 2 August 2026
- Published by Springer Nature
- p. 63-74
BioReFineNet: A Hybrid Dense U-Net ++ Architecture for High-Precision Blood Cell Segmentation from Microscopic Slides
- 7 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 942-947
A Machine Learning–Driven IoT Architecture for Intelligent Irrigation and Fertigation in Horticultural Crops
Effective management of water and nutrients is critical to improving productivity and sustainability in horticultural crop production systems that are experiencing increased climate variability, and limitations in water and nutrient resources. This paper presents an Internet of Things (IoT) architecture driven by Machine Learning (ML) for intelligent irrigation and fertigation management that is specific to horticultural systems. The proposed framework integrates a real-time multi-sensor network of soil moisture, soil temperature, soil electrical conductivity (EC), soil pH, ambient humidity, and vapor-pressure deficit (VPD) sensors with a NodeMCU microcontroller cloud data infrastructure. A supervised learning model based on the Extreme Gradient Boosting (XGBoost) algorithm performs multi-class irrigation decision-making, placing water requirements in three categories: no irrigation, light irrigation, and heavy irrigation. The developed model achieved a testing accuracy of 95.0%, along with additional metrics that provided strong performance benchmarks including a macro-average F1-score of 0.94, and a ROC-AUC score of 0.97, showing high reliability for prediction at the field level. An anomaly detection module provides operational safety by identifying sensor drift, pipeline leakage, and actuator failure to entrain an automated fail-safe mode. The irrigation management system accommodates dynamically changing environmental dynamics by extending the data acquisition period and periodically retraining the model. The experimental research undertaken in agricultural fields of horticulture indicates increased water-use efficiency, decreased fertilizer consumption, and improved nutrient uniformity as opposed to standard irrigation based on rules. Thus, this study illustrates that the coupling of IoT sensing technology with intelligent machine-learning control is possibly disruptive for the future of precision sustainable horticulture.
- 7 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 942-947
A Machine Learning–Driven IoT Architecture for Intelligent Irrigation and Fertigation in Horticultural Crops
Effective management of water and nutrients is critical to improving productivity and sustainability in horticultural crop production systems that are experiencing increased climate variability, and limitations in water and nutrient resources. This paper presents an Internet of Things (IoT) architecture driven by Machine Learning (ML) for intelligent irrigation and fertigation management that is specific to horticultural systems. The proposed framework integrates a real-time multi-sensor network of soil moisture, soil temperature, soil electrical conductivity (EC), soil pH, ambient humidity, and vapor-pressure deficit (VPD) sensors with a NodeMCU microcontroller cloud data infrastructure. A supervised learning model based on the Extreme Gradient Boosting (XGBoost) algorithm performs multi-class irrigation decision-making, placing water requirements in three categories: no irrigation, light irrigation, and heavy irrigation. The developed model achieved a testing accuracy of 95.0%, along with additional metrics that provided strong performance benchmarks including a macro-average F1-score of 0.94, and a ROC-AUC score of 0.97, showing high reliability for prediction at the field level. An anomaly detection module provides operational safety by identifying sensor drift, pipeline leakage, and actuator failure to entrain an automated fail-safe mode. The irrigation management system accommodates dynamically changing environmental dynamics by extending the data acquisition period and periodically retraining the model. The experimental research undertaken in agricultural fields of horticulture indicates increased water-use efficiency, decreased fertilizer consumption, and improved nutrient uniformity as opposed to standard irrigation based on rules. Thus, this study illustrates that the coupling of IoT sensing technology with intelligent machine-learning control is possibly disruptive for the future of precision sustainable horticulture.
- 12 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 763-768
Enhanced Meningitis Type Classification Using Ensemble Learning: A Hybrid LightGBM Model with Advanced Preprocessing
Effective patient management and treatment depend on the timely and accurate diagnosis of meningitis types, such as bacterial, viral, and fungal. Using a hybrid model based on LightGBM within an ensemble learning paradigm, this research suggests an improved machine learning framework for meningitis type classification. The approach includes a strong preprocessing pipeline to manage the complexity of clinical data in the real world. In addition to mean imputation for missing numerical values and StandardScaler for feature normalization, this pipeline incorporates Label Encoding for all categorical features, including “Gender,” “Pathogen_Present,” “Outcome,” and “Risk_Level.” A LightGBM classifier, which is intended to function as a central element within a larger hybrid ensemble architecture, is then trained using the preprocessed data. The model's strong discriminatory capabilities are demonstrated through evaluation on a specific test set. 92% accuracy, 91% precision, and 92% recall were attained by the LightGBM classifier. An F1-Score of $\mathbf{0. 9 1 0 0}(\mathbf{9 1 \%})$ is one of the additional performance metrics. These findings highlight the value of combining sophisticated preprocessing methods with a potent gradient boosting model, such as LightGBM, to improve meningitis classification diagnostic accuracy and open the door to more dependable clinical decision support systems.- 12 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 763-768
Enhanced Meningitis Type Classification Using Ensemble Learning: A Hybrid LightGBM Model with Advanced Preprocessing
Effective patient management and treatment depend on the timely and accurate diagnosis of meningitis types, such as bacterial, viral, and fungal. Using a hybrid model based on LightGBM within an ensemble learning paradigm, this research suggests an improved machine learning framework for meningitis type classification. The approach includes a strong preprocessing pipeline to manage the complexity of clinical data in the real world. In addition to mean imputation for missing numerical values and StandardScaler for feature normalization, this pipeline incorporates Label Encoding for all categorical features, including “Gender,” “Pathogen_Present,” “Outcome,” and “Risk_Level.” A LightGBM classifier, which is intended to function as a central element within a larger hybrid ensemble architecture, is then trained using the preprocessed data. The model's strong discriminatory capabilities are demonstrated through evaluation on a specific test set. 92% accuracy, 91% precision, and 92% recall were attained by the LightGBM classifier. An F1-Score of $\mathbf{0. 9 1 0 0}(\mathbf{9 1 \%})$ is one of the additional performance metrics. These findings highlight the value of combining sophisticated preprocessing methods with a potent gradient boosting model, such as LightGBM, to improve meningitis classification diagnostic accuracy and open the door to more dependable clinical decision support systems.- 25 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 290-295
A Hybrid Model for Meningitis Diagnostic Accuracy with Ensemble Learning: Combining Logistic Regression and L-BFGS
Prompt and accurate diagnosis of meningitis, a severe inflammation of the meninges, is vital for effective patient care. This research introduces a novel diagnostic method designed to improve the precision of meningitis detection, utilizing Logistic Regression enhanced by the Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimization technique. Leveraging the robust classification power of Logistic Regression, refined through L-BFGS, aims to achieve superior diagnostic outcomes. Essential to the methodology is the incorporation of Cerebrospinal Fluid (CSF) analysis, a key element in meningitis diagnosis, alongside clinical symptoms, laboratory results, and patient demographics within the comprehensive dataset. The model showed strong results, achieving 94.25% accuracy and a 93.65% F1-score. Should it have been computed, the area under the ROC curve would have further detailed its performance. Findings demonstrate that Logistic Regression, optimized with L-BFGS and incorporating CSF data, significantly improves diagnostic accuracy when compared to standard Logistic Regression models and those without CSF information. This research highlights the potential of L-BFGS-enhanced Logistic Regression, especially with crucial diagnostic data like CSF analysis, to improve the diagnosis of complex medical conditions such as meningitis. This suggested technique provides a beneficial instrument for improving patient treatment and potentially lessening the effects of incorrect diagnoses on healthcare organizations. Subsequent research will center on verifying the model's validity using larger, more varied data collections, investigating the inclusion of alternative machine learning methodologies, and additionally examining the precise influences of cerebrospinal fluid data on diagnostic precision, with the intention of maximizing clinical usefulness.
- 25 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 290-295
A Hybrid Model for Meningitis Diagnostic Accuracy with Ensemble Learning: Combining Logistic Regression and L-BFGS
Prompt and accurate diagnosis of meningitis, a severe inflammation of the meninges, is vital for effective patient care. This research introduces a novel diagnostic method designed to improve the precision of meningitis detection, utilizing Logistic Regression enhanced by the Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) optimization technique. Leveraging the robust classification power of Logistic Regression, refined through L-BFGS, aims to achieve superior diagnostic outcomes. Essential to the methodology is the incorporation of Cerebrospinal Fluid (CSF) analysis, a key element in meningitis diagnosis, alongside clinical symptoms, laboratory results, and patient demographics within the comprehensive dataset. The model showed strong results, achieving 94.25% accuracy and a 93.65% F1-score. Should it have been computed, the area under the ROC curve would have further detailed its performance. Findings demonstrate that Logistic Regression, optimized with L-BFGS and incorporating CSF data, significantly improves diagnostic accuracy when compared to standard Logistic Regression models and those without CSF information. This research highlights the potential of L-BFGS-enhanced Logistic Regression, especially with crucial diagnostic data like CSF analysis, to improve the diagnosis of complex medical conditions such as meningitis. This suggested technique provides a beneficial instrument for improving patient treatment and potentially lessening the effects of incorrect diagnoses on healthcare organizations. Subsequent research will center on verifying the model's validity using larger, more varied data collections, investigating the inclusion of alternative machine learning methodologies, and additionally examining the precise influences of cerebrospinal fluid data on diagnostic precision, with the intention of maximizing clinical usefulness.
- 30 January 2025
- Published by Springer Nature
- p. 203-217
Forecasting Heart Disease Using Deep BI-DI Neural Networks
- 30 January 2025
- Published by Springer Nature
- p. 203-217
Forecasting Heart Disease Using Deep BI-DI Neural Networks
- 13 November 2024
- Published by Taylor & Francis
- p. 213-223
A Study on Machine Learning Twig Graphs on the Hyper Wiener Index of Complete Graph
A straightforward connected undirected graph GTG has a Wiener index WTG(G) equal to half of the distance between each pair of vertices. According to hyper-Wiener index HWTG(G), half of the distance between all pairs of vertices in GTG is squared. In this chapter, the authors present and discuss HWTG(G) for Kn⊙(Pn⊙K1) and Kn⊙Tn graphs. In graph theory, HWTG(G) extends the classical WTG(G). In a graph, it computes the sum of raised-to-a-power distances between vertices. Machine learning techniques can be applied to analyze and predict properties of graphs based on the hyper-Wiener index, similar to its application in other graph indices. In addition, the authors provide C++ algorithms to compute HWTG(G) for these graphs.
- 13 November 2024
- Published by Taylor & Francis
- p. 213-223
A Study on Machine Learning Twig Graphs on the Hyper Wiener Index of Complete Graph
A straightforward connected undirected graph GTG has a Wiener index WTG(G) equal to half of the distance between each pair of vertices. According to hyper-Wiener index HWTG(G), half of the distance between all pairs of vertices in GTG is squared. In this chapter, the authors present and discuss HWTG(G) for Kn⊙(Pn⊙K1) and Kn⊙Tn graphs. In graph theory, HWTG(G) extends the classical WTG(G). In a graph, it computes the sum of raised-to-a-power distances between vertices. Machine learning techniques can be applied to analyze and predict properties of graphs based on the hyper-Wiener index, similar to its application in other graph indices. In addition, the authors provide C++ algorithms to compute HWTG(G) for these graphs.
- 28 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1260-1264
Enhancing Meningitis Type Diagnosis through a Novel Hybrid Ensemble of Bagging and Gradient Boosting
Meningitis, an inflammation of the meninges surrounding the brain and spinal cord, remains a significant public health concern globally. Accurately diagnosing different types of meningitis is crucial for providing appropriate treatment and improving patient outcomes. In this study propose a novel hybrid ensemble approach that combines the strengths of bagging and gradient boosting algorithms to enhance the diagnosis of meningitis types. The bagging technique involves training multiple decision tree models on random subsets of the training data and aggregating their predictions to reduce over fitting and improve generalization. By integrating these two powerful ensemble methods, this approach aims to capture complex patterns in the data and provide robust and accurate predictions. The proposed hybrid ensemble was evaluated on a comprehensive dataset of meningitis patients, including clinical symptoms, laboratory tests, and demographic information. The model's performance was assessed using various metrics such as accuracy, F1-score, and area under the receiver operating characteristic (ROC) curve. Our results demonstrate that the hybrid ensemble significantly outperformed individual bagging and gradient boosting models, as well as other commonly used algorithms like logistic regression and support vector machines. The optimized hyper parameters and selected features provide insights into the most important factors influencing meningitis types, which can aid clinicians in making informed decisions. This study highlights the potential of hybrid ensemble methods in enhancing the diagnosis of complex medical conditions like meningitis. Combining bagging and gradient boosting can lead to better precision, F1 score, and recall compared to using them separately, the values are 0.94,0.95. 0.96 respectively. The proposed approach can be easily adapted to other disease classification tasks and serves as a valuable tool for improving patient care and reducing healthcare costs associated with misdiagnosis. Future research should focus on validating the model's performance on larger and more diverse datasets and exploring the integration of additional machine learning techniques to further enhance diagnostic accuracy.
- 28 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1260-1264
Enhancing Meningitis Type Diagnosis through a Novel Hybrid Ensemble of Bagging and Gradient Boosting
Meningitis, an inflammation of the meninges surrounding the brain and spinal cord, remains a significant public health concern globally. Accurately diagnosing different types of meningitis is crucial for providing appropriate treatment and improving patient outcomes. In this study propose a novel hybrid ensemble approach that combines the strengths of bagging and gradient boosting algorithms to enhance the diagnosis of meningitis types. The bagging technique involves training multiple decision tree models on random subsets of the training data and aggregating their predictions to reduce over fitting and improve generalization. By integrating these two powerful ensemble methods, this approach aims to capture complex patterns in the data and provide robust and accurate predictions. The proposed hybrid ensemble was evaluated on a comprehensive dataset of meningitis patients, including clinical symptoms, laboratory tests, and demographic information. The model's performance was assessed using various metrics such as accuracy, F1-score, and area under the receiver operating characteristic (ROC) curve. Our results demonstrate that the hybrid ensemble significantly outperformed individual bagging and gradient boosting models, as well as other commonly used algorithms like logistic regression and support vector machines. The optimized hyper parameters and selected features provide insights into the most important factors influencing meningitis types, which can aid clinicians in making informed decisions. This study highlights the potential of hybrid ensemble methods in enhancing the diagnosis of complex medical conditions like meningitis. Combining bagging and gradient boosting can lead to better precision, F1 score, and recall compared to using them separately, the values are 0.94,0.95. 0.96 respectively. The proposed approach can be easily adapted to other disease classification tasks and serves as a valuable tool for improving patient care and reducing healthcare costs associated with misdiagnosis. Future research should focus on validating the model's performance on larger and more diverse datasets and exploring the integration of additional machine learning techniques to further enhance diagnostic accuracy.
- 28 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 159-164
Cross-Language Connect: A Chat App for Smooth Multilingual Communication
The rise of social media has firmly established chat as a cornerstone of online interaction. However, language barriers continue to hinder effective communication among users from diverse linguistic backgrounds. Many traditional chat platforms restrict users to a single language, often requiring the use of external translation tools for cross-lingual conversations, a process that can be cumbersome and inefficient. To address this challenge, a groundbreaking chat application designed to shatter language barriers is proposed. By eliminating the need for a shared language, the solution enables seamless communication between individuals from various linguistic backgrounds. Existing chat platforms frequently confine users to one language, necessitating external translation tools for multilingual interactions. It introduces a cutting-edge chat application that seamlessly integrates machine translation into the chat experience, removing the need for additional applications and facilitating unrestricted multilingual communication. To support this vision, an advanced Neural Machine Translation (NMT) model has been developed that effectively addresses the complexities of cross-lingual communication. The model outperforms commercial alternatives such as Google and Bing Translators in terms of BLEU scores (BLEU-1: 0.87 vs. 0.91/0.83; BLEU-2: 0.77 vs. 0.81/0.73), demonstrating superior fluency and accuracy. Additionally, the model achieves a competitive F1 score of 0.87 compared to Google Translator's 0.92, indicating its strong performance in capturing the nuances of translated text. This technological advancement empowers the chat application to deliver seamless and precise multilingual conversations, enabling global connections and enhancing cultural understanding. Through this innovative approach, a future where language serves as a bridge connecting people from all corners of the world, rather than a barrier that divides them, is envisioned.
- 28 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 159-164
Cross-Language Connect: A Chat App for Smooth Multilingual Communication
The rise of social media has firmly established chat as a cornerstone of online interaction. However, language barriers continue to hinder effective communication among users from diverse linguistic backgrounds. Many traditional chat platforms restrict users to a single language, often requiring the use of external translation tools for cross-lingual conversations, a process that can be cumbersome and inefficient. To address this challenge, a groundbreaking chat application designed to shatter language barriers is proposed. By eliminating the need for a shared language, the solution enables seamless communication between individuals from various linguistic backgrounds. Existing chat platforms frequently confine users to one language, necessitating external translation tools for multilingual interactions. It introduces a cutting-edge chat application that seamlessly integrates machine translation into the chat experience, removing the need for additional applications and facilitating unrestricted multilingual communication. To support this vision, an advanced Neural Machine Translation (NMT) model has been developed that effectively addresses the complexities of cross-lingual communication. The model outperforms commercial alternatives such as Google and Bing Translators in terms of BLEU scores (BLEU-1: 0.87 vs. 0.91/0.83; BLEU-2: 0.77 vs. 0.81/0.73), demonstrating superior fluency and accuracy. Additionally, the model achieves a competitive F1 score of 0.87 compared to Google Translator's 0.92, indicating its strong performance in capturing the nuances of translated text. This technological advancement empowers the chat application to deliver seamless and precise multilingual conversations, enabling global connections and enhancing cultural understanding. Through this innovative approach, a future where language serves as a bridge connecting people from all corners of the world, rather than a barrier that divides them, is envisioned.
- 15 May 2024
- Published by Springer Nature
- p. 507-516
An Optimized Diagnostic Precision Using Genetic Algorithm in Breast Cancer
- 15 May 2024
- Published by Springer Nature
- p. 507-516
An Optimized Diagnostic Precision Using Genetic Algorithm in Breast Cancer