Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
10 results
Search Results
Item Dialect Identification Using Spectral and Prosodic Features on Single and Ensemble Classifiers(Springer Verlag, 2018) Chittaragi, N.B.; Prakash, A.; Koolagudi, S.G.In this paper, investigation of the significance of spectral and prosodic behaviors of speech signal has been carried out for dialect identification. Spectral features such as cepstral coefficients, spectral flux, and entropy are extracted from shorter frames. Prosodic attributes such as pitch, energy, and duration are derived from longer frames. IViE (Intonational Variations in English) speech corpus covering nine dialectal regions of British Isles has been considered, to evaluate the proposed approach. Since corpus is available in both read and semi-spontaneous modes, the influence of spectral and prosodic behavior over these datasets is distinguishably articulated. Further, two distinct classification algorithms, namely support vector machine (SVM) and an ensemble of decision trees along with the SVM are used for identification of nine dialects. Dialect discriminating information captured from both features are used for constructing feature vectors. Experiments have been conducted on individual and combinations of features. A better dialect recognition performance is observed with ensemble methods over a single independent SVM. © 2017, King Fahd University of Petroleum & Minerals.Item Stacking Deep learning and Machine learning models for short-term energy consumption forecasting(Elsevier Ltd, 2022) Sujan Reddy, A.; Akashdeep, S.; Harshvardhan, R.; Kamath S․, S.Accurate prediction of electricity consumption is essential for providing actionable insights to decision-makers for managing volume and potential trends in future energy consumption for efficient resource management. A single model might not be sufficient to solve the challenges that result from linear and non-linear problems that occur in electricity consumption prediction. Moreover, these models cannot be applied in practice because they are either not interpretable or poorly generalized. In this paper, a stacking ensemble model for short-term electricity consumption is proposed. We experimented with machine learning and deep models like Random Forests, Long Short Term Memory, Deep Neural Networks, and Evolutionary Trees as our base models. Based on the experimental observations, two different ensemble models are proposed, where the predictions of the base models are combined using Gradient Boosting and Extreme Gradient Boosting (XGB). The proposed ensemble models were tested on a standard dataset that contains around 500,000 electricity consumption values, measured at periodic intervals, over the span of 9 years. Experimental validation revealed that the proposed ensemble model built on XGB reduces the training time of the second layer of the ensemble by a factor of close to 10 compared to the state-of-the-art, and also is more accurate. An average reduction of approximately 39% was observed in the Root mean square error. © 2022 Elsevier LtdItem Spatial Mapping of Flood Susceptibility Using Decision Tree-Based Machine Learning Models for the Vembanad Lake System in Kerala, India(American Society of Civil Engineers (ASCE), 2023) Kulithalai Shiyam Sundar, P.; Kundapura, S.Floods have claimed the lives of countless people and caused significant property damage in many countries, putting their livelihoods in the jeopardy. The Vembanad lake system (VLS) in Kerala, India, has faced adverse mishappening during 2018, 2019, and 2021 floods in the state due to torrential rainfall. The goal of this research is to construct effective decision tree-based machine learning models such as adaptive boosting (AdaBoost), random forest (RF), gradient boosting machines (GBMs), and extreme gradient boosting (XGBoost) for integrating data, processing, and generating flood susceptibility maps. There are 18 conditioning parameters considered, which include seven categories and 11 numerical data. These seven categorical data were converted to numerical data, bringing the total amount of input data to 61. The recursive feature elimination (RFE) was utilized as the feature selection technique, and a total of 22 layers were chosen to feed into the machine learning models to generate the flood susceptibility maps. The efficiencies of the models were evaluated using receiver operating characteristic (ROC)-area under the ROC curve (AUC), F1 score, accuracy, and kappa. According to the results, the performance of all four models demonstrated their practical application; however, XGBoost fared well in terms of the model's metrics. For the testing data set, the ROC-AUC values of XGBoost, GBM, and AdaBoost are 0.90, whereas it was 0.89 for RF. The accuracy varied significantly among the four models, with XGBoost scoring 0.92, followed by GBM (0.88), RF (0.87), and AdaBoost (0.87). As a result, this map may be utilized for early mitigation actions during future floods, as well as for land-use planners and emergency managers, assisting in the reduction of flood risk in regions prone to this hazard. © 2023 American Society of Civil Engineers.Item InFLuCs: Irradiance Forecasting Through Reinforcement Learning Tuned Cascaded Regressors(IEEE Computer Society, 2024) Chandrasekar, A.; Ajeya, K.; Vinatha Urundady, U.Accurate prediction of solar irradiance is essential for optimizing renewable energy sources in distributed generation systems due to its significant impact on solar power generation. Despite notable advancements, the inherent variability of irradiance presents challenges for existing models. In this article, we introduce a novel approach for irradiance forecasting using a cascaded combination of regressors applied to transformed process variables. Our method utilizes a gradient-boosted decision tree as the primary regressor to generate initial predictions, which are subsequently refined by a support vector regressor acting as an error correction module. Notably, the secondary regressor's kernel, alongside other hyperparameters, is dynamically learned through reinforcement learning with an RNN-based controller. Evaluation results demonstrate that our prediction-correction framework achieves superior performance compared to state-of-the-art approaches, as indicated by RMSE, MAE, and text{R}^{2} score metrics. Thorough comparative analysis highlights the model's enhanced accuracy and its potential for precise irradiance forecasting. © 2005-2012 IEEE.Item Exploring and understanding the microwave-assisted pyrolysis of waste lignocellulose biomass using gradient boosting regression machine learning model(Elsevier Ltd, 2024) Sinha, S.; Sankar Rao, C.; Kumar, A.; Venkata Surya, D.; Basak, T.The production of bio-oil is a complex process influenced by various parameters. Optimizing these parameters can significantly enhance bio-oil yield, thus improving process efficiency. This study aims to develop a predictive model for bio-oil yield using the Gradient Boosting Regression (GBR) technique. It also seeks to identify the key factors affecting bio-oil yield and determine the optimal conditions for maximizing production. The GBR model was constructed using data collected from the literature. The model's performance was evaluated based on its determination coefficients for training and testing datasets. Optimization studies were conducted to identify the best conditions for bio-oil production. The GBR model demonstrated high precision, with determination coefficients of 0.983 and 0.913 for the training and testing datasets, respectively, indicating its effectiveness in predicting bio-oil yield. The optimal conditions for maximizing bio-oil yield were identified as 20 min of pyrolysis time, a temperature of 771 °C, and 524W of microwave power. The two-way PDP analysis provided valuable insights into the interactive effects of temperature with other factors, enhancing the understanding of the dynamics of the bio-oil production process. This study not only identifies the most impactful variables for bio-oil yield but also offers critical guidance for optimizing the production process. © 2024 Elsevier LtdItem Multiscenario Analysis of Hydrological Responses to Climate Change over River Basins of the Western Ghats of India(American Society of Civil Engineers (ASCE), 2024) Shetty, S.; Umesh, P.; Shetty, A.In the face of rising greenhouse gas concentrations, our study investigates the intricate regional dynamics of hydrological responses across three vital river basins of the Western Ghats of India. Employing advanced eXtreme Gradient Boosting (XGBoost) ensemble models based on Coupled Model Intercomparison Project (CMIP6) data, the article explores the anticipated changes in the climate variables under two future scenarios. The findings reveal a compelling narrative of temperature fluctuations, with increased warming in future decades from November to June ushering in warmer winters and extended summer seasons. These climatic shifts carry profound implications for rainfall patterns, potentially disrupting rainfall during the pivotal months of June and July up to the decade 2030s, with a more pronounced increase in the Purna River Basin (PRB) after the decade 2050s. The projected future climate scenarios indicate that the Vamanapuram River Basin (VRB) and PRB will experience contrasting patterns of dry and wet events, with the VRB facing severe to extreme dry and the PRB witnessing increased moderate to extreme wet events under high-emission scenarios. Additionally, the PRB may experience the paradox of increasing wetness and aridity. These insights provide crucial guidance for policy formulation and adaptation measures to safeguard agriculture and other vital sectors in the face of evolving climate conditions. © 2024 American Society of Civil Engineers.Item A hybrid model of convolutional neural network and an extreme gradient boosting for reliability evaluation in composite power systems integrated with renewable energy resources(Springer Science and Business Media Deutschland GmbH, 2025) Yarramsetty, C.; Moger, T.; Jena, D.This paper introduces an approach that enhances the computational efficiency of reliability assessment for composite power systems by integrating machine learning (ML) techniques with sequential monte carlo simulation (SMCS). Integration of renewable energy resources (RERs) into power systems is increasing at a rapid pace. Evaluating the reliability of composite power systems is helpful in identifying any deficiencies in their operation. As power systems operation becomes more fluctuating and stochastic, it is necessary to update the tools used to analyse reliability. In this paper, SMCS is used as a conventional method, as it provides results by taking chronological nature of RERs. However, SMCS is highly computational. ML models fit for solving complex problems that require computational power. ML techniques, such as convolutional neural network (CNN) and hybrib models of Convolutional and Extreme Gradient Boosting (ConXGB), and Convolutional and Random Forest (ConRF) are proposed to determine the expectation of load curtailment and minimum amount of load curtailments. The proposed technique is applied on test system IEEE RTS-79. Results indicate the ConvXGB method is fast and accurate in computing composite reliability indices. For instance, it achieved a Loss of Load Probability (LOLP) of 0.0025 and an Expected Demand Not Supplied (EDNS) of 0.1850 MW, compared to SMCS’s LOLP of 0.0021 and EDNS of 0.1794 MW while reducing computational time from 12900 to 5414 s. These results confirm the proposed method’s speed and accuracy, making it a robust solution for modern power system reliability evaluation. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.Item An integrated frequency domain decomposition and deep neural network approach for short-term PV power forecast(Springer Science and Business Media Deutschland GmbH, 2025) Kumar, A.; Kashyap, Y.; Rai, A.Weather disturbances and atmospheric parameters significantly influence the fluctuations in PV power output, which in turn affect the stability of grid operations. The current study proposed short-term PV power forecasting based on appropriate cutoff frequency in frequency domain and artificial intelligence method. Initially, the actual PV power data are decomposed into the frequency domain, and optimal cutoff frequency is determined by minimizing the squared difference of correlation between the decomposed components. Subsequently, the PV power is separated into low-frequency components (LFC) and high-frequency components (HFC). Then, long short-term memory (LSTM) and light gradient boosting machine (LGBM) models are then employed to forecast the LFC and HFC PV power. The final forecast output is generated using various recombination methods. The proposed combined forecast model, LFC-LGBM + HFC-LGBM, based on frequency domain decomposition (FDD) and LGBM approach, demonstrates superior performance compared to models (LFC-LSTM + HFC-LSTM), (LFC-LGBM + HFC-LSTM), and (LFC-LSTM + HFC-LGBM). The best-performing model (LFC-LGBM + HFC-LGBM) achieves a MAE of 4.9420%, a RMSE of 7.1047%, and a correlation index (R) of 0.9734 for 15-min ahead timesteps. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.Item Leveraging explainable AI framework for predictive modeling of products of microwave pyrolysis of lignocellulosic biomass using machine learning(Elsevier B.V., 2025) Kale, R.D.; Lenka, M.; Sankar Rao, C.The accurate prediction of biochar, bio-oil, and biogas yields in biomass pyrolysis is critical for optimizing process efficiency and sustainable biofuel production. In this study, machine learning (ML) models were developed using literature-derived data on biomass composition and pyrolysis conditions to predict product yields. A comparative analysis of multiple ML algorithms revealed that Decision Tree and Extra Trees exhibited the highest predictive accuracy, followed by Random Forest, Gradient Boosting Trees, and Extreme Gradient Boosting. LightGBM, Gaussian Process Regression, and CatBoost provided moderate performance, while AdaBoost demonstrated the lowest accuracy. To enhance interpretability, Explainable Artificial Intelligence (XAI) techniques, specifically SHAP analysis, were employed to identify key factors influencing pyrolysis yields. Temperature, ash content, fixed carbon, and moisture content were the dominant parameters governing biochar yield, whereas heating rate, reaction time, and feedstock properties such as carbon and volatile matter content significantly influenced bio-oil production. The gas yield was primarily driven by temperature, with secondary cracking mechanisms enhancing non-condensable gas formation. These insights provide a data-driven foundation for optimizing biomass pyrolysis processes, enabling targeted valorization strategies for lignocellulosic feedstocks. The integration of ML and XAI in this study establishes a transparent and interpretable modeling framework, facilitating informed decision-making for sustainable biofuel production. © 2025 Elsevier B.V.Item Predicting wave reflection coefficient of vertical caisson breakwater using machine learning: A data-driven approach(Elsevier Ltd, 2025) Shankara Krishna, A.; Rao, M.; Rao, S.Coastal zones are vital for ecological balance and human development, but are increasingly threatened by wave activity, shoreline erosion, and sea-level rise. To mitigate these challenges, engineers employ coastal protection structures. Specifically, vertical caisson breakwaters are preferred in deeper waters due to their advantages. Reflection Coefficient is an important hydrodynamic performance indicator for breakwaters. Recently, machine learning (ML) has been used for predicting coastal engineering parameters, offering an efficient means to support or augment traditional physical model studies, particularly during preliminary design phases, if sufficient quality data is available. This research focuses on using ML models to estimate the reflection coefficient of vertical caisson breakwaters based on a limited set of experimental data. Four different algorithms- Artificial Neural Network (ANN), Random Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGB)- are developed and evaluated. Hyperparameters are optimised using a hybrid approach, combining Grid Search with manual refinement. Of the four models, XGB achieved the highest prediction accuracy (Test CC = 0.9631), while Random Forest exhibited the smallest generalisation gap, indicating strong consistency across datasets. The findings from the study suggest that XGB offers an efficient tool for early-stage design applications in coastal engineering. © 2025 Elsevier Ltd