Search

Total electron data (TEC) from GPS nowadays can be used as a tool for understanding the space weather phenomena. The development of prediction model for TEC is quiet crucial and challenging due to the dynamic behavior of the ionosphere, since it depends on different factors such as seasonal, diurnal and spatial variations, solar geomagnetic conditions etc. In this paper, an attempt is made for predicting the GPS derived TEC values for different GNSS stations over India using a hybrid method based on Ensemble empirical mode decomposition (EEMD) and Long Short-Term Memory (LSTM) deep learning method. The daily TEC time series data from the IISc Bangalore (Latitude 13.021, Longitude 77.570), Lucknow (Latitude 26.912, Longitude 80.956) and Hyderabad (Latitude 17.417, Longitude 78.551) stations over India during the period 2008 to 2015 of solar cycle 23 and 24 is used for analysis. The assessment of model performance for testing predicted output compared with LSTM and EMD-LSTM models, and their comparison results show that the hybrid EEMD-LSTM model presents better than the other models. [ABSTRACT FROM AUTHOR]

As the demand for renewable energy is increasing, wind speed forecasting (WSF) becomes increasingly popular and essential for wind power generation. Several methods have been developed in the literature for WSF. However, WSF accuracy is challenging to obtain due to its non-stationary and non-linear character. To overcome this issue, this study suggests a novel integrated forecasting model for WSF. In the majority of cases, decomposition techniques are essential for removing noise and extracting patterns from non-stationary and non-linear wind speed time series. Among these decomposition techniques, the most widely used decomposition methods based on frequency domain theory are the Fourier decomposition method (FDM), the empirical wavelet transform (EWT), and the variational mode decomposition (VMD). However, FDM decomposition gives inconsistent results, EWT has mixed problems, and VMD is unsuitable for non-stationary time series. Thus, a novel combined model based on empirical Fourier decomposition (EFD) techniques, long short-term memory (LSTM) neural networks, and the Grey Wolf optimizer (GWO) is proposed in this paper. Initially, the wind speed time series is decomposed using EFD into a set of sub-series and a residual, which is a stationary sub-series that can be easily modelled by a recurrent neural network (RNN). Further, the matching LSTM is used to forecast each sub-series and residual, while the GWO optimizes the sub-series estimated output, resulting in improved prediction precision and stability. In order to assess the model performance, five wind speed datasets from various places in India are employed. The findings show that the proposed model is the top performing model among the other ten models; The average statistical score obtained from all datasets reduces 11.01% in the mean absolute error (MAE), 10.68% in the root mean square error (RMSE) and 10.73% in the mean absolute percentage error (MAPE). Further, regularity conditions for universal consistency of proposed model is also proved mathematically. • A novel hybrid EFD-LSTM-GWO architecture for wind speed forecasting. • Resolving the mixed-mode, inconsistent, and non-stationary challenges. • Analysing the proposed model's universal consistency. • Outperformance of the proposed algorithm over other existing models. [ABSTRACT FROM AUTHOR]

Early rainfall prediction is very important to ensure the economic balance of any agriculture-dominated country, such as India. Deep learning has recently received considerable attention for predicting rainfall using historical data. However, rainfall depends on multiple factors like ElNino, rising sea levels, warming of the Atlantic Ocean, etc. The changing climate has also significantly increased rainfall variability. The current deep-learning approaches are either based on a limited set of predictors or do not utilize the time-series properties for the recurrence of similar climate conditions every 12 months. Besides, they perform poorly in predicting accurate rainfall estimates for the diverse areas of India due to limited data availability across various regions. Therefore, this paper presents a groundbreaking solution to address the challenge of limited data by introducing RfGan (Rainfall GAN), a novel variant of generative adversarial networks. RfGan aims to fulfill the crucial need for abundant data to ensure the effectiveness of model training. Moreover, a novel rainfall prediction method, RfGanNet, has been introduced that leverages the strengths of the proposed RfGan for data augmentation, long-short-term memory, and convolutional neural networks for rainfall prediction. The performance of RfGanNet has been validated for India and its clustered regions in terms of nine parameters: precision, recall, specificity, Matthews correlation coefficient, accuracy, F1-score, mean absolute error, mean square error and computation time. The experimental results of RfGan are compared against original and synthetic datasets generated from state-of-the-art methods, namely tabular gan (TGAN), Wasserstein gan (WGAN), and conditional tabular gan (CTGAN). The experimental results indicate that the proposed RfNet model, when trained using RfGan-generated data, surpassed all the other considered methods by achieving 97.22 % accuracy. Moreover, RfNet has been compared with other related deep Learning approaches to validate its efficacy. Additionally, an ablation study investigating various LSTM–CNN combinations has been conducted. The findings of the study indicate that oceanic and climate-related predictors have a noteworthy influence on Indian rainfall. Moreover, the study highlights the suitability of combining Bidirectional LSTM, CNN, and multiple predictors for accurate rainfall prediction. • Novel variant of generative adversarial network- RfGan is proposed. • Different combinations of LSTM and CNN for rainfall prediction are tested. • Novel classifier — RfNet is tested on different augmented datasets. • K-means clustered regions of India have been identified. [ABSTRACT FROM AUTHOR]

• An automated glaucoma detection methodology is proposed. • Use of features like GIST and PHOG for capturing the textural and shape properties. • The results are evaluated on publicly available database named Drishti-GS1 and HRF. • The paper focuses on accuracy and AUC parameters to evaluated the performance. • The method works better than existing methods implemented on Drishti-GS1 and HRF database. Effective diagnosis of glaucoma mainly relies on the analysis of optic disc characteristics of retina. Glaucoma is considered as second leading cause of blindness and its early detection prevents patients from temporary or permanent blindness. It effects the intensity and shape near optic disc of the retina. Fundus photography has revolutionized the field of ophthalmology and helped in visualizing the structure of optic disc. The proposed work aims to develop an automated diagnostic system based on fundus images for glaucoma disease. It focuses on extraction of GIST and pyramid histogram of oriented gradients (PHOG) features from preprocessed fundus images. The extracted features are ranked and selected through principal component analysis (PCA) to choose significant features. The classification into glaucomatous images is done with SVM classifier on fundus images of Drishti-GS1 and HRF databases. The results obtained from the proposed method are compared with recent glaucoma detection techniques in the literature, including deep learning methodologies, on the basis of accuracy and AUC parameters. The performance of the system is also validated by glaucoma expert from Sharp Sight Group of Eye Hospitals, Delhi-NCR, India. [ABSTRACT FROM AUTHOR]

Ground-based multi-frequency microwave radiometer Profiler (Microwave Radiometer Profiler-MRP) provides valuable information on the altitude distribution of atmospheric humidity and temperature in the troposphere (upto ∼ 10 km) with high temporal resolution (of ∼ 1 min) under all-weather condition. These information are vital inputs for characterization of the atmospheric boundary layer, convective cloud systems, local weather modification and for studying the atmospheric dynamics. Potential of such radiometers are well established over the mid-latitude region. Thiruvananthapuram, located in the Indian peninsular region, is one of the first stations in the equatorial region where such an instrument made continuously operational for seven years since 2010 at the tropical coastal station. The instrument measures the brightness temperatures (T b in Kelvin) at Ka- and V-band frequencies with a temporal resolution of ∼ 1 min. A trained back-propagation neural network technique based on the radiosonde ascends data for more than one decade obtained by India Meteorological station at Thiruvananthapuram, was used to derive the atmospheric humidity and temperature profiles. This paper presents the implementation of another retrieval technique based on deep learning approach - batch normalization and neural network (BNN) to retrieve temperature and water vapour density (WVD) profiles from T b values during clear sky conditions and validation of the retrieval accuracies. This inversion model, much faster in computation, consists of two hidden layers and the rectified linear unit (ReLU) is used as the activation function as it can overcome the problems of saturation and vanishing gradients. The T b s observed by radiometer are corrected by reducing the bias between the simulated T b , using forward radiative transfer model, and the observed T b. The validation of retrieved profiles with the radiosonde demonstrates a good retrieval capability, showing a root-mean-square error of 1.8 K for temperature, < 2 g/m 3 for WVD and ∼ 14 % for RH. • The bias between simulated and observed brightness temperature has been estimated. • Neural network with two hidden layers and ReLU activation layer has been used.. • Retrieved profiles are validated using radiosonde data and exhibit improved accuracy. [ABSTRACT FROM AUTHOR]

Highlights • The paper deals with two different water quality indicators for the Bakreswar reservoir. • The main indicators chosen here are dissolved oxygen and zooplankton abundance. • Artificial neural network modelling has been used to make predictions of DO and zooplankton abundance. • ANN are capable of accurately predicting the dependent variable based on predictor variables. • ANN modelling is better suited in the current scenario over traditional approaches like multiple regression. Abstract Presence of optimal levels of dissolved oxygen (above critical level of 4.5 mg L−1) and presence of zooplankton community are indicators of good water quality of an aquatic ecosystem and also of the health of the same. Reservoirs being artificially created water bodies present hybrid systems containing features of both lotic and lentic systems and thus have unique organization that are representative of both rivers and lakes. Since any reservoir is primarily a fresh water system, presence of a large array of zooplankton (diverse community structure) implies its good health and also presence of optimal dissolved oxygen levels supports sustenance of life. In this study, artificial neural network modelling approach has been utilized to predict the level of dissolved oxygen and zooplankton abundance in the Bakreswar reservoir and their variation in relation to the environmental factors. Use of neural network modelling is exceedingly capable of determining correlation among apparently non correlated environmental data and in the current study these are capable of accurately predicting the variations in the levels of dissolved oxygen as well as the abundance of zooplankton. From this study, it has been observed that chemical factors like productivity, nitrates, salinity, pH , phosphates, total dissolved solids, etc. are mostly responsible for control of dissolved oxygen and zooplankton variation at certain points of the study site (stations 1 and 3) whereas at other points (station 2) physical factors like solar radiation, humidity, etc. are more effective. These models are capable of finding the important environmental controllers of such variations and prove to be a powerful alternative to traditional approaches like multiple regression analysis. [ABSTRACT FROM AUTHOR]

• A systematic approach for optimal portfolio construction based on sector analysis is presented. • The concepts of probability theory, econometrics, and deep learning are integrated. • Evidence theory and Granger causal network are unified to select strong and diverse economic sectors. • Deep recurrent neural network is used for optimal ratio allocation to portfolio securities. • The proposed model is designed for investment in National Stock exchange of India. With the emerging areas of economy, the diverse sector-based investment portfolios are considered more significant. This paper presents an integrated approach of portfolio construction based on sector analysis for investment in National Stock Exchange of India. The model is developed in four stages: First, multiple evidences are collected using historical data to analyze the performance of different sectors, and ranked using Dempster-Shafer evidence theory. The top-ranked sectors are considered as the strong sectors. Second, Granger causality test is employed for determining interdependencies between the strong sectors using the past closing price time series. Then, the strong sectors with no causal relationships are selected to diversify the investment. Third, leading stocks of the selected sectors are picked on the basis on their performance in previous years. Thus, the portfolio of best stocks from strong and diverse sectors is constructed. Fourth, for optimizing of the constructed portfolio or to obtain optimal ratio allocations of stocks in the portfolio, an optimization function is constructed and simulated using a deep recurrent neural network. The results obtained by the proposed portfolio are found very significant which confirms its effectiveness. The applicability of the proposed model is verified by empirical study over the existing models. [ABSTRACT FROM AUTHOR]

Day-ahead load forecasting plays a crucial role in operation and management of power systems. Weather conditions have a significant impact on daily load profile, hence, it follows an almost similar pattern within a season. However, it varies markedly across the seasons. Existing literature on load forecasting adopts a very casual approach in considering the seasonality, based on either calendar month or some meteorological parameter, which is inconsistent and inaccurate, especially during transition periods, thus leading to high forecasting errors. This paper proposes a novel seasonal segmentation approach for day-ahead load forecasting that uses multiple bidirectional Long Short Term Memory (LSTM) networks. An index has been derived from various weather parameters that govern the selection of seasonal models for forecasting purposes. Weighted output from multiple seasonal models is also possible in special cases for better forecasting accuracy. The proposed seasonal segmentation approach avoids the need for frequent model retraining and results in a better forecast accuracy with a relatively simple LSTM structure. The performance of the proposed method has been validated and compared on the actual load data of Madhya Pradesh state (MP), India. The improved results suggest that the proposed approach can be applied reliably for load scheduling applications. • An encoder-decoder Bi-LSTM model with attention is developed to train the models. • An adaptive method for accurate identification of seasonal duration is proposed. • An index based selective and weighted method is developed for optimized forecasts. • Dynamic Time Warping (DTW) is used to adaptively select the forecast models. [ABSTRACT FROM AUTHOR]

Multimodality has shown to be helpful in several natural language processing tasks. Thus, adding multiple modalities to the traditional sentiment analysis has also proven to be useful. However, multimodality in a low resource setting for sentiment analysis is yet to be explored for several resource-constrained languages. Assamese is a low-resource language spoken mainly in the state of Assam in India. This paper presents an Assamese multimodal dataset comprising of 16,000 articles from the news domain as a benchmark resource. Secondly, we present a multi-stage multimodal sentiment analysis framework that concurrently exploits textual and visual cues to determine the sentiment. The proposed architecture encodes the news content collaboratively. The text branch encodes semantic content information by considering the semantic information of the news. At the same time, the visual branch encodes the visual appearance information from the news image. Then, an intermediate fusion-based multimodal framework is proposed to exploit the internal correlation between textual and visual features for joint sentiment classification. Finally, a decision-level fusion mechanism is employed on the three models to integrate cross-modal information effectively for final sentiment prediction. Experimental results conducted on the Assamese dataset built in-house demonstrate that the contextual integration of multimodal features delivers better performance (89.3%) than the best unimodal features (85.6%). • A multimodal and multi-purpose Assamese news dataset is presented. • Two models are proposed for visual and textual modality for sentiment classification. • A multi-stage fusion approach is employed for multimodal sentiment classification. • Qualitative and quantitative analysis are carried out of the models. [ABSTRACT FROM AUTHOR]

To combat the recent coronavirus disease 2019 (COVID-19), academician and clinician are in search of new approaches to predict the COVID-19 outbreak dynamic trends that may slow down or stop the pandemic. Epidemiological models like Susceptible–Infected–Recovered (SIR) and its variants are helpful to understand the dynamics trend of pandemic that may be used in decision making to optimize possible controls from the infectious disease. But these epidemiological models based on mathematical assumptions may not predict the real pandemic situation. Recently the new machine learning approaches are being used to understand the dynamic trend of COVID-19 spread. In this paper, we designed the recurrent and convolutional neural network models: vanilla LSTM, stacked LSTM, ED_LSTM, BiLSTM, CNN, and hybrid CNN+LSTM model to capture the complex trend of COVID-19 outbreak and perform the forecasting of COVID-19 daily confirmed cases of 7, 14, 21 days for India and its four most affected states (Maharashtra, Kerala, Karnataka, and Tamil Nadu). The root mean square error (RMSE) and mean absolute percentage error (MAPE) evaluation metric are computed on the testing data to demonstrate the relative performance of these models. The results show that the stacked LSTM and hybrid CNN+LSTM models perform best relative to other models. • Various variants of LSTM and CNN deep learning architectures have been designed. • Predicted COVID-19 confirmed cases for India and its four most affected states. • RMSE and MAPE are computed to compare the performance of the designed models. • Experimental results show that predicted cases are very close to actual cases. [ABSTRACT FROM AUTHOR]

Social Networking Sites have significant potential to reveal valuable explicit as well as implicit statistics and patterns when deep learning is applied to their raw and unstructured data. Tweets posted by the users on their timeline not only reflect their mindset, their likes and dislikes but could also be used to unveil significant amount of information about many psychological aspects and behavior that may be hard to study directly. This paper aims to predict the personality of the 100 real-time Twitter users conforming to personality traits in the BIG 5 model by extracting features from their tweets using ensemble of CNN (Convolutional Neural Network) and BiLSTM (Bidirectional Long Short Memory). The findings of our experiment shows that our model performs slightly better than previous baselines methods achieving an accuracy 75.134% on testing data. We have further hypothesized that unrestricted data available on Twitter may contain features that can be used to predict the personality of its user. It was concluded that personality of Twitter users in the real world is reflected in their online behaviour, reinforcing the premise that the nature of online interactions does not significantly differ from that of real-world interactions. Overall, the study provides a deep insight into the impact of social media data in providing predictive indicators of user behavior. • Twitter data can provide predictive indicators of user behavior. • We hypothesized personality of an individual is highly correlated with their online behavior. • The study aims to predict the personality of Indian Twitter users. • Deep learning ensemble model has been employed for the study. • It is observed that nature of online interactions does not differ from real-world interactions. [ABSTRACT FROM AUTHOR]

Tropical cyclone prediction is essential to limit death toll and damage caused by them. In this paper, a model has been formulated to classify the cyclone as no cyclone, minimal, moderate, extensive, extreme and catastrophic in view of attributes such as Wind Speed(10 m), Rainfall, Wind Direction(10 m), Sea Surface Temperature(2 m), Sea Level Pressure(2 m) and Relative humidity(2 m). The models have been trained using the meteorological data from MERRA-2 Web service which provides time-series data with spatial resolution of approximately 50 km. The cyclone data is taken from RSMC — New Delhi for Tropical cyclone disturbances over Indian Oceans. Initially, the models are trained using deep learning networks like MLP, LSTM, GRU, RNN, BI-LSTM and CNN. Since CNN gives better results, the CNN model is chosen for further analysis. The hyper parameters of the CNN model are optimized using genetic algorithm. The values drawn from genetic algorithm appear to be promising than the values which were chosen manually in random. The model is then modified by removing the fully connected layer which operates as a classifier in CNN network. The conventional machine learning classifiers like — Decision Tree, K-Nearest neighbor, logistic regression, Naive Bayes, Random Forest, SVM and XGBoost are used as a classifier in the place of a fully connected layer in CNN. Further to increase the prediction accuracy, C4.5 Decision tree algorithm is modified to be used as a classifier in CNN. Classification is performed by considering the Spatio-temporal data of various important cities in India. The model was tested to classify the category for 5 different cyclones and was also compared with Saffir–Simpson's scale to validate the correctness of the model. The proposed model gives a better performance compared to the conventional machine learning and deep learning classifiers in terms of time complexity, accuracy, precision and recall and this can supplement the cyclone prediction process of current NWP models. • Cyclone Prediction is proposed using CNN, modified C4.5 and genetic algorithms. • CNN is used for feature extraction and genetic algorithm for hyper parameter tuning. • Modified C4.5 is proposed to solve the limitations of C4.5 decision tree algorithm. • Accuracy of the proposed model is significantly improved compared to the existing works. [ABSTRACT FROM AUTHOR]