Browsing by Author "S. Karthikeyan"

Now showing 1 - 13 of 13

A Comparative Analysis for Forty Years of Land Use Land Cover Change (1991–2021) Using Cart and Random Forest Classifiers for Varanasi District (India)
(Springer, 2025) Annu Kumari; S. Karthikeyan
Effective implementation of remote sensing images using classification techniques facilitates us with the extracting spatial and temporal data classification of LULC. Mapping interchanges in LULC pattern is paving a way for investigating the influence of several socio-economic and environmental factors on the surface of Earth. Our case study depicts a comparison analysis between two algorithms using Landsat temporal images to analyze the interchanges in the class of LULC for the Varanasi district of India. We applied Random Forest (RF) Classification and Classification and Regression Tree (CART) classification in our case study, in the Google Earth Engine platform (GEE) extracting images from Landsat 5, Landsat 7 and Landsat 8 since 1991—2021 period. We evaluated the performance metrics with auxiliary data and several spectral indices on our final classification accuracy on both the classifiers. We used multi-spectral Landsat bands’ (Landsat 7– Landsat 8) of spatial resolution from 30 to 15 m. The results shows: 1) values of spectral indices applied Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-Up Index (NDBI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), Normalized Difference Water Index (NDWI), Moisture Soil Adjusted Vegetation Index (MSAVI), Soil Water Index (SWI), Normalized Difference Moisture Index (NDMI), Normalized Burn Ratio (NBR), Normalized Burn Ratio2 (NBR2), Urban Index (UI), Visible Red Based Built-up Index (VrNIR_BI), Visible Green Based Built-up Index (VgNIR_BI), 2) Accuracy assessment 98.3%, 97.37%, 96.48%, 95.3% of RF and 87.14%, 91.45%, 89.23%, 88.67% of CART for Training accuracy, training kappa, testing accuracy and testing kappa respectively. Resulting, Random Forest classier outperforms well as compared to CART classifier in our case study. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.
Combining GRNN and SVM using Receiver Operating Characteristics (ROC) for improved classification of non coding RNA
(2012) P.K. Singh; S. Karthikeyan
The RNAs are by and large coded in to protein, but there are big classes of RNA's, which are not being coded in to protein. These RNA are called non coding RNA (ncRNA) and are important in regulatory, catalytic, or structural roles in the cell. The ncRNAs are prevalent in all living organisms. An automated method is beneficial for classification of these ncRNAs from the rest. In this paper we are proposing a classifier fusion method to improve the classification performance which is better than performance of the stand alone classifiers. We have combined two Classifiers namely SVM and GRNN under Receiver Operating Characteristic (ROC) space using the maximum likelihood combination. The input features are generated from the basic geometric and topological properties of RNA secondary structure. Then we have tested a set of sequences which is not seen (unknown) by GRNN and SVM classifiers during training process with cross-fold verification. The results obtained by the fusion method shows better than that were individually obtained by GRNN and SVM. © 2012 IEEE.
Comparative Performance of Maximum Likelihood and Minimum Distance Classifiers on Land Use and Land Cover Analysis of Varanasi District (India)
(Springer Science and Business Media Deutschland GmbH, 2023) Annu Kumari; S. Karthikeyan
Monitoring the Land Use/Land Cover (LULC) changes in this present era has become the most demanding task and it is very crucial for planning, proper resource management, regulating the expansion in the fringe areas in the existing cities, etc. Now the collection of reliable data has become easier as the hyperspectral images captured by various remote sensing satellites are readily available in different spatial and spectral resolutions. Processing this data according to the Region of Interest (ROI) in order to extract meaningful information is challenging. The most crucial part of this process is to identify various land covers very accurately. Only an automatic classification provides a feasible solution, as the manual process is tedious, expensive, and time-consuming. This paper compares two different image classification algorithms in classifying the covers, which can be utilized for land use and land cover changing pattern analysis in the Varanasi district of India. The experiments were carried out with the two most popular classification algorithms, namely: The Maximum Likelihood classifier and the Minimum Distance classifier. The overall accuracy and kappa co-Efficient values computed are 41.67 and 0.12 for the Minimum Distance Classifier and 82.43 and 0.78 for the Maximum Likelihood Classifier. It has been observed that the Maximum Likelihood Classifier outperforms the Minimum Distance Classifier. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
Dual nature hidden layers neural networks "A novel paradigm of neural network architecture"
(2001) S. Karthikeyan; Ravi Prakash; B.B. Paul
We present here a new scheme to construct a neural network architecture based on the physiological properties biological neuron for enhancing its performance. The new scheme divides every hidden layer into two parts to facilitate the processing of 0 and 1 separately and reduces the total number of interconnections considerably. The first part consist of units that receives signals only from '1 state' units of the immediately lower layer and are responsible for producing excitation units in the output layer i.e. the '1 state' and the second part consist of units that also receives signals only '1 state' units from the immediately lower layer are responsible for producing inhibition of the units in the output layer i.e. the '0 states'. The resulting architecture converges fast, produces more reliable results and reduces the computational burden considerably when compared to fully connected neural networks.
Enhanced identification of β-lactamases and its classes using sequence, physicochemical and evolutionary information with sequence feature characterization of the classes
(Elsevier Ltd, 2017) Abhigyan Nath; S. Karthikeyan
β-lactamases provides one of the most successful means of evading the therapeutic effects of β lactam class of antibiotics by many gram positive and gram negative bacteria. On the basis of sequence identity, β-lactamases have been identified into four distinct classes- A, B, C and D. The classes A, C and D are the serine β-lactamases and class B is the metallo-lactamse. In the present study, we developed a two stage cascade classification system. The first-stage performs the classification of β-lactamases from non-β-lactamases and the second-stage performs the further classification of β-lactamases into four different β–lactamase classes. In the first-stage binary classification, we obtained an accuracy of 97.3% with a sensitivity of 89.1% and specificity of 98.0% and for the second stage multi-class classification, we obtained an accuracy of 87.3% for the class A, 91.0% for the class B, 96.3% for the class C and 96.4% for class D. A systematic statistical analysis is carried out on the sieved-out, correctly-predicted instances from the second stage classifier, which revealed some interesting patterns. We analyzed different classes of β-lactamases on the basis of sequence and physicochemical property differences between them. Among amino acid composition, H, W, Y and V showed significant differences between the different β-lactamases classes. Differences in average physicochemical properties are observed for isoelectric point, volume, flexibility, hydrophobicity, bulkiness and charge in one or more β-lactamase classes. The key differences in physicochemical property groups can be observed in small and aromatic groups. Among amino acid property group n-grams except charged n-grams, all other property group n-grams are significant in one or more classes. Statistically significant differences in dipeptide counts among different β-lactamase classes are also reported. © 2017 Elsevier Ltd
Enhanced Prediction and Characterization of CDK Inhibitors Using Optimal Class Distribution
(International Association of Scientists in the International Association of Scientists in the, 2017) Abhigyan Nath; S. Karthikeyan
Cyclin-dependent kinase inhibitors (CDKIs) govern the regulation of cyclin-dependent kinases, which are responsible for controlling cell cycle progression. The members of the CDKI protein family play important roles in many processes like tumor suppression, apoptosis, transcriptional regulation. The sequence similarity-based search methods to annotate putative CDKIs do not yield optimal performance due to sequence diversity of CDKIs. As a consequence, machine learning-based models have become viable choices for predicting CDKI. In this work, we have developed a framework for handling the class imbalance factor (which is encountered very frequently in biological datasets) in order to enhance the prediction of CDKI through machine learning approaches. We have designed our experiments to achieve the optimal performance of machine learning-based methods in predicting CDKI by investigating the dataset-related prediction enhancement issues, like: (1) What should be the optimal class distribution ratio in the training set? (2) Should we oversample or undersample? (3) At what ratio, positive and negative samples should be oversampled or undersampled? and (4) How to select the best-performing classifier? We have addressed these issues through comparing the results from an imbalanced training set with training sets which are created at different resampling rates by using synthetic minority over-sampling technique and undersampling technique to have varied class distributions. The proposed framework resulted in 100 % sensitivity, 93.7 % specificity, 96.4 % accuracy, 0.929 MCC with 0.981 AUC using simple sequence-based features on a leave-one-out cross-validation test. The generalization ability of the trained model was further tested on four separate blind testing sets. Our work supports the fact that the performance of the algorithms can be enhanced by creating an optimal class distribution in the training set besides fine-tuning of the parameters of the algorithms. This optimal ratio of positive and negative samples in the training set is an important learning enhancement parameter for prediction models based on machine learning algorithms. © 2016, International Association of Scientists in the Interdisciplinary Areas and Springer-Verlag Berlin Heidelberg.
Enhanced prediction of recombination hotspots using input features extracted by class specific autoencoders
(Academic Press, 2018) Abhigyan Nath; S. Karthikeyan
In yeast and in some mammals the frequencies of recombination are high in some genomic locations which are known as recombination hotspots and in the locations where the recombination is below average are consequently known as coldspots. Knowledge of the hotspot regions gives clues about understanding the meiotic process and also in understanding the possible effects of sequence variation in these regions. Moreover, accurate information about the hotspot and coldspot regions can reveal insights into the genome evolution. In the present work, we have used class specific autoencoders for feature extraction and reduction. Subsequently the deep features that are extracted from the autoencoders were used to train three different classifiers, namely: gradient boosting machines, random forest and deep learning neural networks for predicting the hotspot and coldspot regions. A comparative performance analysis was carried out by experimenting on deep features extracted from different sets of the training data using autoencoders for selecting the best set of deep features. It was observed that learning algorithms trained on features extracted from the combined class specific autoencoder out performed when compared with the performances of these learning algorithms trained with other sets of deep features. So the combined class-specific autoencoder based feature extraction can be applied to a growing range of biological problems to achieve superior prediction performance. © 2018 Elsevier Ltd
Heuristic Technique to Find Optimal Learning Rate of LSTM for Predicting Student Dropout Rate
(Springer Science and Business Media Deutschland GmbH, 2024) Anuradha Kumari Singh; S. Karthikeyan
Predictive analytics is being increasingly recognized as being important for evaluating university students’ academic achievement. Utilizing big data analytics particularly students’ demographic information, offers valuable insights to bolster academic success and enhance completion rates. For instance, learning analytics is a vital element of big data within university settings, offering strategic decision-makers the chance to conduct time series analyses on learning activities. We have used semesters first and second records of Polytechnic Institute of Portalegre students. Advanced deep learning methods, such as the Long short-term memory (LSTM) model are employed to analyze students at risk of retention issues. Typically, the best design for a deep neural network model was found by trial and error, which is a laborious and exponential combinatorial challenge. Hence we proposed the heuristic technique to configure the parameters of the neural network. The parameter taken into account in this work is the learning rate of the Adam optimizer. In this study, we have presented the ant colony optimization (ACO) technique to determine the ideal learning rate for model training. Experimental results obtained with the predictive model indicated that prediction of student retention is possible with a high level of accuracy using ACO-LSTM approach. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.
Impact of Dimensionality Reduction on Membership Privacy of CNN Models
(Springer Science and Business Media Deutschland GmbH, 2023) Ashish Kumar Lal; S. Karthikeyan
Dimensionality reduction is an essential tool for exploratory data analysis, classification and clustering, manifold learning, and preprocessing in deep learning. The curse of dimensionality is a well-recognized serious problem in machine learning. Many researchers have attempted to improve machine learning accuracy and performance with dimensionality reduction. Nowadays, machine learning models’ privacy vulnerability is an essential quality measure. The effect of dimensionality reduction on the privacy leakage of deep learning models is an understudied research area. This work explored the effects of dimensionality reduction on the privacy leakage of the deep learning model. The experiments for image classification using CNN model were performed on the widely used Cifar10 dataset. The results show that although the PCA technique improves the classification task, it does not enhance the model’s privacy when tested against the membership inference attack. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
Impact of hardware connectivity on Grover’s algorithm in NISQ era
(Springer, 2025) Mohit Joshi; Manoj Kumar Mishra; S. Karthikeyan
The quantum search operation as dictated in Grover’s landmark paper had been a crucial area in the study of quantum algorithms. It has become a critical component in many quantum cryptography and computation algorithms and threatens today’s AES security infrastructure. The quadratic speedup provided by Grover’s algorithm is hampered severely due to the presence of a realistic environment. Many studies have analyzed the effect of different noises on Grover’s search algorithm. However, the efficiency of the algorithm also depends on the connectivity of qubits on realistic quantum hardware. This study evaluated the performance of Grover’s algorithm with varying qubit connectivity under the presence of two-qubit depolarizing noise and single-qubit amplitude damping and dephasing noise. Unidirectional and bidirectional variants of nine coupling maps for qubit connectivity were chosen. The analysis has shown that the transpilation efficiency for Grover’s algorithm is deeply sensitive to the connectivity and degree of the hardware, which influences the depth of the circuit. This, in turn, has a measurable effect on the performance of the algorithm on a particular hardware. This study also ranks the favorable coupling maps using the decision-making technique of AHP-TOPSIS. The analysis has shown that grid, hex, and modified star are the most favorable hardware connectivity. The unidirectional linear, ring, star, and full-connected are the worst choices. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.
Leveraging Grover’s Algorithm for Quantum Searchable Encryption in Cloud Infrastructure and its application in AES Resource Estimation
(Springer, 2024) Mohit Joshi; Manoj Kumar Mishra; S. Karthikeyan
Designing efficient techniques to search over encrypted data space has always been an intriguing security challenge, although many solutions based on classical searching methods have been proposed. Grover’s algorithm, a quantum counterpart of searching algorithms, has proven to provide quadratic speedup over any classical search technique on an unsorted database. However, this algorithm is unable to search over encrypted data space. This study proposed an extension of Grover’s algorithm to enable search over encrypted dataspace, allowing clients with limited-capability quantum resources to delegate complex search operations to an untrusted server. The blindness of data in this protocol is achieved by encrypting qubits using Pauli’s rotation gates that maximally mix the outgoing states. The empirical estimation of the overhead of the computation due to the introduction of this technique has been analyzed. This estimate has been used for comparative analysis, showing the efficiency of the proposed protocol. A practical application of the proposed searchable encryption technique has been utilized to estimate the increase in resources needed to carry out a brute-force attack on AES encryption using secure Grover’s algorithm. Furthermore, an extensive experimental analysis of the effect of noise has been studied using four different noise models: amplitude damping, phase damping, depolarizing noise, and bit-flip noise. The investigation provided useful insight into the behavior of the proposed algorithm under noisy conditions and also estimated the tolerance thresholds of the proposed algorithm under different noise models. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.
Recent Trends and Open Challenges in Blind Quantum Computation
(Springer Science and Business Media Deutschland GmbH, 2023) Mohit Joshi; S. Karthikeyan; Manoj Kumar Mishra
Quantum mechanics with radically novel properties: such as superposition, entanglement, and the no-cloning theorem, has begun to open up Quantum Computation beyond the scope of Classical Computation. In recent years, quantum technology has seen tremendous leaps both in academic research and commercial exploration. Quantum computation has started to find applications in many more new domains every day, from disrupting modern cryptography to enabling unconventional security techniques, from quantum chemistry to physical simulations, and from quantum machine learning to financial optimization. This paper first provides the conceptual groundings of quantum computation and explores blind quantum computation, a one-of-its-kind sub-categorization of quantum computing active research. The use of Blind Quantum Computation in Quantum Cryptography is elaborated in detail. Finally concludes with highlights of the applicability of the subject. The paper is an easy-to-follow guide introducing the research trends and open challenges for the new researcher in follow-up on the field. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
Understanding the protein sequence and structural adaptation in extremophilic organisms through machine learning techniques
(Elsevier, 2020) Abhigyan Nath; S. Karthikeyan
Organisms thriving at extremes of environmental surroundings are called extremophiles. Depending upon the different extreme surrounding conditions, the extremophiles are further classified into different classes, for example, thermophiles and psychrophiles based upon extreme temperature conditions. Extremophilic proteins have paved their way into the workbench of many laboratories for possible industrial applications. Molecular analysis of extremophilic protein sequences can provide a wealth of information about their successful structural and functional adaptations to such extremes of environment. Understanding the protein sequence adaptation parameters will facilitate in providing a set of rules to create the rational design of extremophilic proteins as well as in understanding the conversion processes of a mesophilic protein into some other form of extremophilic protein. Advent of affordable next-generation sequencing technologies resulted in swift increase in the amount of fully sequenced genomes of extremophilic organisms, giving rise to a larger datasets that are sufficient enough to be mined to increase the extremophilic knowledgebase. Supervised machine learning algorithms combined with other computational statistical methods are useful in more accurate prediction of extremophilic proteins. In this chapter, we have described the roadmap for developing supervised machine learning–based prediction models followed by statistical analysis for inferring the molecular basis of extremophilic adaptation. © 2020 Elsevier Inc. All rights reserved.