Publications

Publication Alert!!!

Accurate relationship classification in large-scale clinical terminologies such as SNOMED CT is essential for enabling semantic interoperability, clinical decision support, and advanced healthcare analytics. The framework integrates multiple transformer-based classifiers, each trained with class-weighted loss functions to better capture underrepresented relationship types. A probability calibration mechanism is applied to enhance prediction reliability and interpretability. Experimental results demonstrate that the proposed ensemble approach significantly outperforms baseline and single-model architectures across multiple evaluation metrics, particularly in minority classes. This study highlights the effectiveness of combining domain-adapted language models with ensemble learning and calibration techniques for robust ontology relationship classification.

Ontology relationship classification plays a crucial role in structuring biomedical knowledge and supporting intelligent healthcare systems. Traditional approaches often fail to jointly capture semantic textual information and graph-based structural dependencies. In this paper, we propose a hybrid framework that combines BERT-based contextual embeddings with a Relational Graph Convolutional Network (R-GCN) to model both textual semantics and ontology structure. BERT is used to generate rich contextual representations of concept descriptions, while the R-GCN leverages relational edges to learn inter-concept dependencies within the ontology graph. The integrated model effectively captures both local and global information, leading to improved classification performance. The results suggest that combining transformer-based embeddings with graph neural networks provides a powerful approach for knowledge representation in biomedical informatics.

Jounral Publications

An accurate, very short-term load forecasting is a critical aspect of energy management, enabling utilities, grid operators, and energy providers to make informed decisions regarding the future generation, distribution, and allocation of electricity over a short time horizon. The proposed framework involved two pivotal stages in its development: robust feature engineering and forecasting electric load by a Prophet-XGBoost-CatBoost (Pr-XGB-CB) model. Feature engineering involves utilizing multiple sets of historical electricity consumption data with different time lags for analysis and enhancement. Prophet dissects observed or forecasted data into understandable seasonal, trend, and holiday segments, while XGBoost stands out for its speed and effectiveness, especially when dealing with numerous features in a given scenario. To get the ensemble forecast, the Catboost algorithm is utilized. 

Peer-reviewed Conference Publications

Accurate relationship classification in large-scale clinical terminologies such as SNOMED CT is essential for enabling semantic interoperability, clinical decision support, and advanced healthcare analytics. The framework integrates multiple transformer-based classifiers, each trained with class-weighted loss functions to better capture underrepresented relationship types. A probability calibration mechanism is applied to enhance prediction reliability and interpretability. Experimental results demonstrate that the proposed ensemble approach significantly outperforms baseline and single-model architectures across multiple evaluation metrics, particularly in minority classes. This study highlights the effectiveness of combining domain-adapted language models with ensemble learning and calibration techniques for robust ontology relationship classification.

Ontology relationship classification plays a crucial role in structuring biomedical knowledge and supporting intelligent healthcare systems. Traditional approaches often fail to jointly capture semantic textual information and graph-based structural dependencies. In this paper, we propose a hybrid framework that combines BERT-based contextual embeddings with a Relational Graph Convolutional Network (R-GCN) to model both textual semantics and ontology structure. BERT is used to generate rich contextual representations of concept descriptions, while the R-GCN leverages relational edges to learn inter-concept dependencies within the ontology graph. The integrated model effectively captures both local and global information, leading to improved classification performance. The results suggest that combining transformer-based embeddings with graph neural networks provides a powerful approach for knowledge representation in biomedical informatics.



Though wind power capacity all over the world is increasing rapidly, the availability of wind power generation mostly relies on wind speed, which is a random variable with a stochastic nature. Therefore, a robust technique with            powerful feature extraction capability is   required to predict wind speed accurately. In this paper, we have recommended a hybrid model using a convolutional neural network    (CNN) and long-short-term memory (LSTM). where CNN is used for extracting fuzzy input features and LSTM to catch the sequence to predict wind speed accurately. As deep learning models are associated with multiple hyperparameters with great impact, the Bayesian optimization algorithm is used for hyperparameter tuning.


As the production and consumption of electricity are simultaneous, an electric power load forecasting technique with higher accuracy can play a pivotal role in a stable and effective power supply system. In this paper, a multivariate Bayesian optimization-based Long short-term memory (LSTM) neural network is proposed to forecast the residential electric power load for the upcoming hour. The Bayesian optimization algorithm is conducted to select the best-fitted hyperparameter values since deep learning networks are associated with different hyperparameters, which play a vital role in the performance of a network architecture.


Wind power capacity around the world is increasing day by day, but the production of wind energy greatly depends on the wind speed, where the wind speed has a stochastic nature over time. In this paper, an artificial neural network (ANN) technique to forecast wind speed for the next hour in Newfoundland, Canada, is proposed. As deep learning models are combined with different hyperparameters, in our study, the selection of important hyperparameters is conducted by applying the Bayesian optimization algorithm.