Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
18 results
Search Results
Item Prediction of crime hot spots using spatiotemporal ordinary kriging(Springer Verlag service@springer.de, 2019) Deshmukh, S.S.; Annappa, B.Prediction can play a very important role in many types of domains, including the criminal justice system. Even a little information can be gained from proper police assignments, which can increase the efficiency of the crime patrolling system. Citizens can also be aware and alert for possible future criminal incidents. This was identified previously, but the proposed solutions use many complex features, which are difficult to collect, especially for developing and underdeveloped countries, and the maximum accuracy obtained to date using simple features is around 66%. Few of these countries have even started collecting such criminal records in digital format. Thus, there is a need to use simple and minimal required features for prediction and to improve prediction accuracy. In the proposed work, a spatiotemporal ordinary kriging model is used. This method uses not only minimal features such as location, time and crime type, but also their correlation to predict future crime locations, which helps to increase accuracy. Past crime hot spot locations are used to predict future possible crime locations. To address this, the Philadelphia dataset is used to extract features such as latitude, longitude, crime type and time of incident, and prediction can be given for every 0.36 square km per day. The city area is divided into grids of 600 × 600 m. According to the evaluation results, the average sensitivity and specificity obtained for these experiments is 90.52 and 88.63%, respectively. © Springer Nature Singapore Pte Ltd. 2019.Item LATA – Label attention transformer architectures for ICD-10 coding of unstructured clinical notes(Institute of Electrical and Electronics Engineers Inc., 2021) Mayya, V.; Kamath S․, S.S.; Sugumaran, V.Effective code assignment for patient clinical records in a hospital plays a significant role in the process of standardizing medical records, mainly for streamlining clinical care delivery, billing, and managing insurance claims. The current practice employed is manual coding, usually carried out by trained medical coders, making the process subjective, error-prone, inexact, and time-consuming. To alleviate this cost-intensive process, intelligent coding systems built on patients’ structured electronic medical records are critical. Classification of medical diagnostic codes, like ICD-10, is widely employed to categorize patients’ clinical conditions and associated diagnoses. In this work, we present a neural model LATA, built on Label Attention Transformer Architectures for automatic assignment of ICD-10 codes. Our work is benchmarked on the CodiEsp dataset, a dataset for automatic clinical coding systems for multilingual medical documents, used in the eHealth CLEF 2020-Multilingual Information Extraction Shared Task. The experimental results reveal that the proposed LATA variants outperform their basic BERT counterparts by 33-49% in terms of standard metrics like precision, recall, F1-score and mean average precision. The label attention mechanism also enables direct extraction of textual evidence in medical documents that map to the clinical ICD-10 diagnostic codes. © 2021 IEEE.Item Ensemble Learning Approach for Short-term Energy Consumption Prediction(Association for Computing Machinery, 2022) Sujan Reddy, A.; Akashdeep; Harshvardhan; Kamath S․, S.Predicting electricity consumption accurately is crucial for garnering insights and potential trends into energy consumption for effective resource management. Due to the linearity/non-linearity in usage patterns, electricity consumption prediction is challenging and cannot be adequately solved by using a single model. In this paper, we propose ensemble learning based approaches for short-term electricity consumption on an open dataset. The ensemble model is built on the combined predictions of supervised machine learning and deep learning base models. Experimental validation showed that the proposed ensemble model is more accurate and decreases the training time of the second layer of the ensemble by a factor close to ten, compared to the state-of-the-art. We observed a reduction of approximately 34% in the Root mean squared error for the same size of historical window. © 2022 Owner/Author.Item Using Stacking Ensemble Method for Rental Bike Prediction(Springer Science and Business Media Deutschland GmbH, 2025) Akashdeep, S.; Mahalinga, A.N.; Harshvardhan, R.; Chinnahalli KomariGowda, S.; Patil, N.Rental bike platforms that improve mobility comfort are on the rise in major cities worldwide. One of the essential requirements for these rental bike systems is that bikes are available to end users at the specified time, reducing waiting time. Increased waiting time indicates that movement has been halted, implying that more efficiency can be gained. As a result, the city’s main priority is ensuring a steady supply of bicycles. It’s crucial to be able to forecast the number of bikes needed at each hour for this. This work look at alternative models for forecasting the bike count per hour needed to maintain a steady supply of bikes. Weather data (Temperature, Humidity, Wind speed, Dew point), the quantity of bikes hired every hour, and time information are all used to train the models. Filtering can also be used to exclude non-predictive parameters and rank features based on how well they predict outcomes. The effectiveness of the regression model was assessed using a testing set after they had been trained using repeated cross-validation. For the model Gradient Boosting Machine, the optimum R2 value is 0.96. The most significant predictors are also determined, as well as their relationships. Bike-sharing demand, data mining, predictive analytics, public bikes, regression. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.Item Automated microaneurysms detection for early diagnosis of diabetic retinopathy: A Comprehensive review(Elsevier B.V., 2021) Mayya, V.; Kamath S․, S.S.; Kulkarni, U.Diabetic retinopathy (DR), a chronic disease in which the retina is damaged due to small vessel damage caused by diabetes mellitus, is one of the leading causes of vision impairment in diabetic patients. Detection of the earliest clinical sign of the advent of DR is a critical requirement for intervention and effective treatment. Ophthalmologists are trained to identify DR, based on examining specific minute changes in the eye - microaneurysms, retinal haemorrhages, macular edema and changes in the retinal blood vessels. Segmentation of microaneurysms (MA) is a critical requirement for the early diagnosis of DR and has been the primary focus of the research community over the past few years. In this work, a systematic review of existing literature is carried out to examine the diagnostic use of automated MA detection and segmentation for early DR diagnosis. We mainly focus on existing early DR diagnosis techniques to understand their strengths and weaknesses. Though early diagnosis is performed using colour fundus photography, fluorescein angiography or optical coherence tomography angiography images, our study is limited to colour fundus based techniques. The early DR diagnosis methodologies reviewed in this article can be broadly classified into classical image processing, conventional machine learning (ML), and deep learning (DL) based techniques. Though significant progress has been achieved in these three classes of early DR diagnosis, several challenges and gaps still exist, underscoring a considerable scope for the development of fully automated, user-friendly early DR diagnosis and grading systems. We discuss in detail the challenges that need to be addressed in designing such effective, efficient, and robust algorithms for early DR diagnosis systems and also the ample scope for future research in this area. © 2021Item Numerical simulation and prediction model development of multiple flexible filaments in viscous shear flow using immersed boundary method and artificial neural network techniques(IOP Publishing Ltd custserv@iop.org, 2020) Kanchan, M.; Maniyeri, R.Many chemical and biological systems have applications involving fluid-structure interaction (FSI) of flexible filaments in viscous fluid. The dynamics of single- and multiple-filament interaction are of interest to engineers and biologists working in the area of DNA fragmentation, protein synthesis, polymer segmentation, folding-unfolding analysis of natural and synthetic fibers, etc. To perform numerical simulation of the above-mentioned FSI applications is challenging. In this direction, methods like the immersed boundary method (IBM) have been quite successful. We simulate the dynamics of multiple flexible filaments subjected to planar shear flow at low Reynolds number using the finite volume method-based IBM. The governing continuity and Navier-Stokes equations are solved by the SIMPLE algorithm on a staggered Cartesian grid system. The validation of the developed model is done using previous works. The length of the filament, its bending rigidity and fluid shear rate are taken as parametric variables and numerical simulations are carried out. Viscous flow forcing and fractional contraction terms are incorporated so as to effectively categorize filament motion into various deformation regimes. The effects of tumbling motion on the filament migration and recuperative aspects are studied. The mutual interaction of two filaments placed side by side is thus observed. Finally, an artificial neural network model is developed from the IBM simulation results to predict tumbling counts for different filament parameters. © 2020 The Japan Society of Fluid Mechanics and IOP Publishing Ltd.Item FarSight: Long-Term Disease Prediction Using Unstructured Clinical Nursing Notes(IEEE Computer Society, 2021) Gangavarapu, T.; S. Krishnan, G.S.; Kamath S?, S.; Jeganathan, J.Accurate risk stratification using patient data is a vital task in channeling prioritized care. Most state-of-the-art models are predominantly reliant on digitized data in the form of structured Electronic Health Records (EHRs). Those models overlook the valuable patient-specific information embedded in unstructured clinical notes, which is the prevalent medium employed by caregivers to record patients' disease timeline. The availability of such patient-specific data presents an unprecedented opportunity to build intelligent systems that provide exclusive insights into patients' disease physiology. Moreover, very few works have attempted to benchmark the performance of deep neural architectures against the state-of-the-art models on publicly available datasets. This article presents significant observations from our benchmarking experiments on the applicability of deep learning models for the clinical task of ICD-9 code group prediction. We present FarSight, a long-term aggregation mechanism intended to recognize the onset of the disease with the earliest detected symptoms. Vector space and topic modeling approaches are utilized to capture the semantic information in the patient representations. Experiments on MIMIC-III database underscored the superior performance of the proposed models built on unstructured data when compared to structured EHR based state-of-the-art model, achieving an improvement of 19.34 percent in AUPRC and 5.41 percent in AUROC. © 2013 IEEE.Item Novel Stock Crisis Prediction Technique - A Study on Indian Stock Market(Institute of Electrical and Electronics Engineers Inc., 2021) Naik, N.; Mohan, B.R.A stock market crash is a drop in stock prices more than 10% across the major indices. Stock crisis prediction is a difficult task due to more volatility in the stock market. Stock price sell-offs are due to various reasons such as company earnings, geopolitical tension, financial crisis, and pandemic situations. Crisis prediction is a challenging task for researchers and investors. We proposed a stock crisis prediction model based on the Hybrid Feature Selection (HFS) technique. First, we proposed the HFS algorithm to removes the irrelevant financial parameters features of stock. The second is the Naive Bayes method is considered to classify the strong fundamental stock. The third is we have used the Relative Strength Index (RSI) method to find a bubble in stock price. The fourth is we have used moving average statistics to identify the crisis point in stock prices. The fifth is stock crisis prediction based on Extreme Gradient Boosting (XGBoost) and Deep Neural Network (DNN) regression method. The performance of the model is evaluated based on Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Square Error(RMSE). HFS based XGBoost method was performed better than HFS based DNN method for predicting the stock crisis. The experiments considered the Indian datasets to carry out the task. In the future, the researchers can explore other technical indicators to predict the crisis point. There is more scope to improve and fine-tune the XGBoost method with a different optimizer. © 2013 IEEE.Item Influence of Support Vector Regression (SVR) on Cryogenic Face Milling(Hindawi Limited, 2021) Karthik, R.M.C.; Malghan, R.L.; Kara, F.; Shettigar, A.; Rao, S.S.; Herbert, M.A.The paper aims to investigate the processing execution of SS316 in manageable machining cooling ways such as dry, wet, and cryogenic (LN2-liquid nitrogen). Furthermore, "one parametric approach"was utilized to study the influence and carry out the comparative analysis of LN2over dry and LN2over wet machining conditions. Response surface methodology (RSM) is incorporated to build a relationship model among the considered independent variables (spindle speed: (S, rpm), feed rate (F, mm/min), and depth of cut (doc) (D, mm)) and the dependent variable (surface roughness (Ra)). Since there is the involvement of more than one independent variable, the generation of regression equation is "multiple linear regression."Based on the attained coefficient value of the independent variable, the respective impact on surface roughness is identified. The results of comparative analysis of LN2over dry and LN2over wet machining states revealed that LN2 machining yielded better surface finish with up to 64.9%, 54.9% over dry and wet machining, respectively, indicating the benefits of LN2 for achieving better Ra. The benchmark function of the proposed mode hybrid-bias (BNN-SVR) algorithm showcases the propensity to emerge out of the local minimum and coincide with the optimal target value. The performance of the (BNN-SVR) is a prevalent new ability to fetch the partially trained weights from the BNN model into the SVR model, thus leading to the conversion of static learning capability to dynamic capability. The performances of the adopted prediction approaches are compared through a range of attained error deviation, i.e., (RA: 3.95%-8.43%), (BNN: 2.36%-5.88%), (SVR: 1.04%-3.61%), respectively. Hybrid-bias (BNN-SVR) is the best suitable prediction model as it provides significant evidence by attaining less error in predicting Ra. However, SVR surpasses BNN and RSM approaches because of the convergence factor and narrow margin error. © 2021 Rao M. C. Karthik et al.Item BeeM-NN: An efficient workload optimization using Bee Mutation Neural Network in federated cloud environment(Springer Science and Business Media Deutschland GmbH, 2021) Shishira, S.R.; Kandasamy, A.Cloud computing is an extensively implemented technique to handle enormous amount of data as it provides flexibility and scalability features. In an established cloud environment, users process their request to share the data that are stored in it. Under the dynamic cloud environment, multiple requests are processed in a short time, which leads to the problem of resource allocation. Virtual Machines or servers aid the cloud in maintaining the workflow active through proper distribution of resources. However, the accurate workload prediction model is necessary for effective resource management. In the present paper, a novel BeeM-NN framework is proposed through the integration of Workload Neural Network Algorithm (WNNA) and Novel Bee Mutation Optimization Algorithm (NBMOA) for optimized workload prediction in a cloud environment. The proposed model encloses the Fitness Feature Extraction Algorithm initially to extract the feature dataset from Azure public dataset and is provided to train the WNNA. The predicted workloads are optimized with the NBMOA in the cloud. The generated model is tested using the workload data traces from the federated cloud service provider and is evaluated and compared with the existing models. The outcome showed the prediction model achieved an accuracy of 99.98% better than the current models with optimum performance in the consumption of resources and cost. The future work is to use the predicted workloads for scheduling in the cloud. © 2020, Springer-Verlag GmbH Germany, part of Springer Nature.