Machine Learning (ML) such as Artificial Neural Network (ANN), Deep learning, Recurrent Neural Networks (RNN), Alex Net and ResNet can be consider as a broad research direction in the identification and classification of critical diseases. CNN and its particular variant usually named as U-Net Segmentation has made a revolutionary advancement in the classification of medical diseases, specifically retinal diseases. However, because of the feature extraction complexity, U-Net has a significant flaw of high memory and CPU consumption, while moving the whole feature map to the related decoder. Furthermore, it can be concatenated to the unsampled decoder feature map avoids the reusing pooling indices. In this research work, a convolutional neural network (CNN) model is proposed for multi-class classification problem with the efficient use of memory consumption. The proposed model has been evaluated on standard benchmark dataset of Eye Net having 32 classes of retinal diseases. From experimental evaluation it has been concluded that the proposed model performs better in terms of memory management and accuracy. The overall comparison has been performed on the bases of precision, recall, accuracy with different number of epochs and time consumption by each step. Proposed technique achieved an accuracy of 95% on Eye-net dataset.

Agriculture is a critical domain, where technology can have a significant impact on increasing yields, improving crop quality, and reducing environmental impact. The use of renewable energy sources such as solar power in agriculture has gained momentum in recent years due to the potential to reduce the carbon footprint of farming operations. In addition to providing a source of clean energy, solar tracking systems can also be used for remote weather monitoring in the agricultural field. The ability to collect real-time data on weather parameters such as temperature, humidity, and rainfall can help farmers make informed decisions on irrigation, pest control, and other crop management practices. The main idea of this study is to present a system that can improve the efficiency of solar panels to provide constant power to the sensor in the agricultural field and transfer real-time data to the app. This research presents a mechanism to improve the arrangement of a photovoltaic (PV) array with solar power and to produce maximum energy. The proposed system changes its direction in two axes (azimuth and elevation) by detecting the difference between the position of the sun and the panel to track the sun using a light-dependent resistor. A testbed with a hardware experimental setup is designed to test the system’s capability to track according to the position of the sun effectively. In the end, real-time data are displayed using the Android app, and the weather data are transferred to the app using a GSM/WiFi module. This research improves the existing system, and results showed that the relative increase in power generation was up to 52%. Using intelligent artificial intelligence techniques with the QoS algorithm, the quality of service produced by the existing system is improved.

The rapid growth of Internet-enabled applications, such as social media platforms, e-commerce sites, and blogs, has led to a surge in user-generated content. This vast amount of data has made sentiment analysis increasingly valuable. Modern Aspect-Based Sentiment Analysis (ABSA) offers a more detailed approach by identifying sentiment trends related to specific aspects within the text. However, the challenge lies in analyzing reviews that are often short, unstructured, and filled with slang and emotive language, making it difficult to gauge customer opinions accurately. To address these issues, we proposed an effective hybrid approach “RoBERTa-1D-CNN-BiLSTM” for ABSA. Initially, the pre-trained Robustly Optimized BERT approach (RoBERTa) and One Dimensional Convolutional Neural Network (1D-CNN) models are used to extract features at the aspect level from the context of the review, following which classification is performed using Bidirectional Long Short-Term Memory (BiLSTM). The approach is evaluated on three cross-domain standards datasets, yielding an accuracy of 92.33%. The results of the experiments show that it surpasses the current leading methods in sentiment analysis and product recommendation.

The advancement of technology has led to the rise of social media forums and e-commerce platforms, which have become popular means of communication, and people can express their opinions through comments and reviews. Increased accessibility to online feedback helps individuals make informed decisions about product purchases, services, and other decisions. This study used a sentiment analysis-based approach to improve the functionality of the recommendations from user reviews and consider the features (aspects and opinions) of products and services to understand the characteristics and attributes that influence the performance of classification algorithms. The proposed model consists of data preprocessing, word embedding, character representation creation, feature extraction using BiLSTM-CF, and classification using BiGRU. The proposed model was evaluated on different multidomain benchmark datasets demonstrating impressive performance. The proposed model outperformed existing models, offering more promising performance results in recommendations.

Scientific document classification is an important field of machine learning. Currently, scientific document category identification is done manually. There are already defined taxonomies available for categorizing scientific documents, such as the Association for Computing Machinery Computing Classification System (ACM CCS) and Bibsonomy. These taxonomies facilitate authors in the categories of their manuscripts. The incorporation of research work from a variety of domains in the assignment takes on the form of a Multi-Label Classification (MLC). Using MLC, it is possible to assign more than one class to a single document. To address the problem of MLC in its entirety, two distinct methods are used (Problem Transformation and Algorithm Adaptation). The MLC dataset is transformed into one or more single-label datasets through the application of the problem transformation technique. Whereas, a single classifier is modified during the algorithm adaptation process so that it can predict multiple labels. Currently, document classification is done using various techniques in the literature, but none of them paid much attention to the problem of imbalance in Multi-Label Datasets (MLD). However, many effective techniques for dealing with imbalance are available in the literature. The goal of this study is to find an effective technique for balancing datasets before multi-label classification to get better predictions for the classes with fewer instances. Six MLDs, nine transformation techniques and seven classifiers are evaluated in this research work. The proposed research will result in a more accurate recommendation of a research topic for a document. For imbalanced MLDs, LPROS is the best resampling technique using statistical tests. When compared to the other classifiers, the BRkNN classifier is better for MLC. This research will facilitate the classification of documents into their respective classes which can be used by various citation indexes.

High dimensionality is a well-known problem that has a huge number of highlights in the data, yet none is helpful for a particular data mining task undertaking, for example, classification and grouping. Therefore, selection of features is used frequently to reduce the data set dimensionality. Feature selection is a multi-target errand, which diminishes dataset dimensionality, decreases the running time, and furthermore enhances the expected precision. In the study, our goal is to diminish the quantity of features of electroencephalography data for eye state classification and achieve the same or even better classification accuracy with the least number of features. We propose a genetic algorithm-based feature selection technique with the KNN classifier. The accuracy is improved with the selected feature subset using the proposed technique as compared to the full feature set. Results prove that the classification precision of the proposed strategy is enhanced by 3 % on average when contrasted with the accuracy without feature selection.