Expansion of Internet of Things (IoT) technologies has greatly enhanced monitoring and management of energy systems, especially in Hybrid Renewable Energy Systems (HRES). This paper presents an IoT-based HRES smart grid framework with a modified Brain Storm Optimization (BSO) algorithm for routing optimization and an Improved Quantum Key Management (IQKM) is a quantum inspired protocol for better data security. The enhanced BSO algorithm, hosted in the cloud infrastructure, optimizes IoT sensor data routing paths, thus diminishing packet transmission latency and improving the network throughput. In contrast to conventional BSO techniques, the enhancement is through dynamic cluster refinement and adaptive node prioritization, designed specifically for real-time cloud-integrated energy systems. In order to protect sensitive energy transmission information, the IQKM protocol includes strong quantum-aided encryption processes and dynamic key creation. These enhancements directly counter the dangers of man-in-the-middle and replay attacks, which exceed capabilities of standard encryption approaches by facilitating low-latency, quantum-resistant communication between HRES nodes. Both Photovoltaic (PV) and wind-based energy sources are utilized by the system to provide power consistently, with cloud-based analytics and IoT sensors ensuring real-time monitoring. Experimental testing via the Adafruit platform reports a 23% Packet Delivery Ratio (PDR) enhancement and 17% encryption/decryption delay reduction compared to baseline and traditional routing algorithms. Such findings ensure the potential for stable, secure, and scalable grid performance by the proposed system.

In this rapidly progressing technological age, the likelihood of transmitting inadequately secured data in the cloud is high. Over the past several decades, experts in data and information security have tested and experimented with numerous hashing combinations. However, these have proven to be insufficient in preventing the interception and decoding of confidential text during transmission. Therefore, methodologically, in this current research, cryptography and steganography is diversified into using three different encryption algorithms of RSA, AES and LSB with further compressing of the to-be-communicated data, to enable the use of limited space in the cloud, as well as permit fast and quick embedded message transmission. The proposed architecture, EdSri has been implemented and tested against few existing models and found to show improved performance in terms of measured security metrics such as password strength, time between login attempts, login attempt rate, failed login attempt rate, device and browser fingerprinting, and also, measuring compression parameters like structural similar index, compression time, compression ratio, compression speed, bit per pixel, saving percentage, mean squared error, and peak to noise signal ratio, EdSri would hopefully become viable platform for exchanging secured information among the cloud communicators when hosted. 

Reinforcement Learning (RL) is a successful and established Artificial Intelligence (AI) method, particularly with recent groundbreaking progress in Deep Reinforcement Learning (DRL). Reinforcement learning is very well suited for sequential decision-making tasks, wherein a learned agent learns an optimal policy after many interactions with an environment. The present paper examines the application of reinforcement learning for automating screening of literature in academic research, particularly in the fields of computer science and e-learning. Keyword filtering techniques, while predominantly applied, are found to be inflexible as well as unable to capture the dynamic nature of research themes. To overcome such constraints, we present a Deep Q-Network (DQN)-based reinforcement learning model that combines reinforcement learning with the Semantic Scholar API to enhance research paper classification based on dynamically acquired decision rules. The proposed reinforcement learning model was trained and tested with a dataset of 8,934 research papers, accessed by systematic searching. The agent filtered and picked 11 effective papers depending on improved selection criteria like publication date, keyword relevance, and scholarly topic provided. The model iteratively optimizes the decision-making process through reward-based learning and therefore maximizes categorization accuracy over time. Test experiments demonstrate utilization of RL-based suggested framework yields classification accuracy at 91.5%, recall at 86.3%, and precision at 89.7%. A comparison test demonstrates that the approach performs 12.5% better on recall and 9.8% better on accuracy compared to traditional keyword-filtering methods. The finding confirms the power of the model in minimizing false positives and false negatives for screening literature, hence proving the scalability and adaptability of reinforcement learning in managing high academic data. This work offers a scalable, cognitive approach to conducting systematic reviews of literature through the application of reinforcement learning to programmatically execute work in academic research. The work shows the promise of reinforcement learning to further enhance research methodology, make literature reviews more effective, and facilitate more knowledgeable decision-making in fast-changing scientific disciplines. Further research will be focused on incorporating hybrid AI models with multi-agent systems of reinforcement learning for responsiveness and classification enhancements.

Malaria remains a significant global health challenge that affects more than 200 million people each year and disproportionately burdens regions with limited resources. Precise and timely diagnosis is critical for effective treatment and control. Traditional diagnostic approaches, including microscopy and rapid diagnostic tests (RDTs), encounter significant limitations such as reliance on skilled personnel, high costs and slow processing times. Advances in deep learning (DL) have demonstrated remarkable potential. They achieve diagnostic accuracies of up to 97% in automated malaria detection by employing convolutional neural networks (CNNs) and similar architectures to analyze blood smear images. This survey comprehensively reviews deep learning approaches for malaria detection and focuses on datasets, architectures and performance metrics. Publicly available datasets, such as the NIH and Delgado Dataset B are evaluated for size, diversity and limitations. Deep learning models which include ResNet, VGG, YOLO and lightweight architectures like MobileNet are analyzed for their strengths, scalability and suitability across various diagnostic scenarios. Key performance metrics such as sensitivity and computational efficiency are compared with models achieving sensitivity rates as high as 96%. Emerging smartphone-based diagnostic systems and multimodal data integration trends demonstrate significant potential to enhance accessibility in resource-limited settings. This survey examines key challenges and includes bias in the data set, generalization of the model and interpretability to identify research gaps and propose future directions to develop robust, scalable and clinically applicable deep learning solutions for malaria detection.

In the field of human-computer interaction, identifying emotion from speech and understanding the full context of spoken communication is a challenging task due to the imprecise nature of emotion, which requires detailed speech analysis. In the area of speech emotion recognition, various techniques have been employed to extract emotions from audio signals, including several well-established speech analysis and classification methods. Despite numerous advancements in recent years, many studies still fail to consider the semantic information present in speech. Our study proposes a novel approach that captures both the paralinguistic and semantic aspects of the speech signal by combining state-of-the-art machine learning techniques with carefully crafted feature extraction strategies. We address this task using feature-engineering-based techniques, which involve extracting meaningful audio features such as energy, pitch, harmonics, pauses, central momentum, chroma, zero-crossing rate, and Mel-frequency cepstral coefficients (MFCCs). These features capture important acoustic patterns that help the model learn emotional cues more effectively. This work is primarily conducted on the IEMOCAP dataset, a large and well-annotated emotional speech corpus. By framing our task as a multi-class classification problem, we extract 15 features from the audio signal and use them to train five machine learning classifiers. Additionally, we incorporate text-domain features to reduce ambiguity in emotional interpretation. We evaluate our model's performance using accuracy, precision, recall, and F-score across all experiments.

The development of medical-imaging neurology diagnostics regarding brain tumor detection and classification via the Internet of Medical Things (IoMT) is important. This research proposes a comprehensive framework addressing user privacy concerns by embedding brain tumor information in a cover image through Hybrid Watermarking Steganography (HWS) using Compressive Sensing Integer Wavelet Transform (CSIWT). The watermarked images are securely transmitted over the IoMT, ensuring data integrity. An Inverse CSIWT-HWS system extracts the hidden brain tumor image for diagnosis. The proposed framework incorporates an Optimized Brain Tumor Segmentation and Classification Network (OBTSC-Net) to enhance diagnostic capabilities. This transfer learning model utilizes Attention Generative Adversarial Networks (AGAN) to segment brain tumor areas from the extracted images, Hybrid Greylag Goose Optimization Genetic Algorithm (HGGO-GA) for disease-specific feature extraction from segmented images, and Broad Learning System Neural Network (BLS-NN) for the accurate classification of benign and malignant brain tumors using BraTS-2020 and BraTS-2021 datasets, offering a reliable and secure tool for remote diagnosis. Finally, the proposed HWS-CSIWT method achieved an average Peak Signal-to-Noise Ratio (PSNR) improvement of 12.65% over existing state-of-the-art methods. The proposed AGAN method achieved an average segmentation accuracy (SACC) improvement of 5.63% over existing methods, and the proposed OBTSC-Net achieved an average classification accuracy (CACC) improvement of 2.82% over existing state-of-the-art methods, confirming its enhanced diagnostic capability in brain tumor classification.

In this paper, we aim to develop a car price prediction model using data collected from an online sales platform. To accomplish the proposed objective, we applied the following approaches and techniques: (1) Collecting sales data from the online sales platform; (2) Exploratory analysis of data before and after data preprocessing; (3) Experimenting to find a suitable prediction model for the collected dataset. The novelty of this study lies in constructing a real-world dataset of pre-owned car prices collected directly from an online sales platform and in building a car price prediction model using an empirical approach combined with machine learning models. Unlike previous studies based on existing structured datasets, this study emphasizes the discovery of data-driven insights through exploratory analysis and the identification of key variables affecting car prices. At the same time, essential insights regarding car prices were obtained from the dataset. Experimental results show that the model using the XGBoost algorithm achieved an R2 of 0.776 for the default parameter case and an R2 of 0.779 for the optimized parameter case. These findings provide a practical solution for real-world car price prediction systems, allowing buyers and sellers to make more informed pricing decisions.

Cognitive distortion refers to the patterns of negative thinking which can distort a person’s perception of reality. These distorted thoughts lead to unhealthy behaviors, emotional distress, and mental health issues, like depression and anxiety. In order to detect cognitive distortion, Deep Learning (DL) techniques are employed; however, these approaches lead to a high error rate and poor performance. This is mainly because they fail to understand the hierarchical semantics, subtle emotional tones, and long-range dependencies within the text. Hence, a new model termed Hierarchical Attention Neural Harmonic Fusion Network (HAN-HFNet) is exploited for cognitive distortion detection from text. Initially, the input sentence is passed to Bidirectional Encoder Representations from Transformers (BERT) tokenization, which generates context-aware embeddings capable of capturing subtle emotional nuances, long-range dependencies, and hierarchical semantics critical for identifying cognitive distortions in text. Next, various Key Performance Indicators (KPIs), like Severity of Cognitive Distortions (SCD), Frequency of Cognitive Distortion (FCD), Correlation Between Cognitive Distortions and Depression Severity, Cognitive Behavioral Therapy (CBT), self-reports of cognitive distortions from individuals, Long-Term Monitoring of Cognitive Distortions (LT-MCD), and impact on daily functioning is considered. Lastly, the cognitive distortion is detected utilizing HAN-HFNet, which is obtained by integrating Hierarchical Deep Learning for Text classification (HDLTex) and Deep High-order Attention neural Network (DHA-Net) using harmonic analysis. This fusion enables the model to learn both coarse and fine-grained features, enhancing contextual understanding and reducing error. Moreover, the performance of the HAN-HFNet is evaluated using the Faulty Information Processing Dataset (FIPD), and it computed a minimum classification error of 0.072, and maximum recall, accuracy, precision, and F1-score of 94.756%, 92.754%, 91.866%, and 93.289%. Furthermore, the model is suitable for integration into real-world mental health support systems, offering scalability and potential deployment in online therapy platforms, clinical decision-making tools, and cognitive behavioral assessment frameworks.

Lung cancer is a main reason of death globally, and reducing death rates and enhancing treatment results depend heavily on quick identification. However, medical image diagnosis, including Computed Tomography (CT) scans, is difficult and demands a high level of experience. This research proposes a comprehensive and interpretable Computer-Aided Diagnosis (CAD) structure to identify lung cancer from medical images. The workflow initiates with an Adaptive Savitzky-Golay Filter, effectively enhancing image quality by smoothing while preserving critical structural edges. Hierarchical Adaptive Cluster Refinement (HACR) is then used for precise segmentation, adaptively identifying abnormal lung regions with high accuracy. For feature extraction, the proposed system utilizes the Deep Statistical Gray-Level Co-occurrence Matrix (DS-GLCM) approach, which captures deep spatial and statistical texture features essential for distinguishing cancerous tissue. At last stage, classification is performed using a novel Deep Learning (DL) model Crested Porcupine Optimized (CPO) Channel-Attention (CA) InceptionResNet.  The CPO algorithm is exploited to tune the CA- InceptionResNet model's hyperparameters. To ensure transparency and reliability in clinical use, Explainable AI (XAI) technique- Local Interpretable Model-Agnostic Explanations (LIME) is used for visual interpretability, highlighting regions in CT images that contribute the most to model forecasts, thus boosting clinician trust and decision-making. The entire framework is implemented in Python, and experimental results on benchmark lung cancer imaging datasets demonstrate its superior performance in terms of performance metrics with an accuracy of 98.18% with sensitivity of 95.94 % and specificity of 99.10%. The combination of advanced DL and explainable AI makes the proposed framework a promising solution for lung cancer diagnosis.

In the context of the growth of information exchange in social networks, messengers, and chats, the problem of spreading disinformation and coordinated inauthentic behaviour by users is becoming increasingly relevant and poses a threat to the state's information security. Traditional manual monitoring methods are ineffective due to the scale and speed of information dissemination, necessitating the development of intelligent automated countermeasures. The paper proposes a hybrid information technology for the automatic detection of disinformation, its sources of spread, and inauthentic behaviour among chat users, combining methods of natural language processing (NLP), machine learning, stylistic and linguistic analysis of texts, and graph analysis of social interactions. Within the study, datasets of authentic and fake messages were compiled, and mathematical models and algorithms for identifying disinformation sources were developed using metrics of graph centrality, clustering, and bigram Laplace smoothing. 
Experimental studies using TF-IDF, BERT, MiniLM, ensemble methods, and transformers confirmed the effectiveness of the proposed approach. The achieved accuracy in disinformation classification is up to 89.5%, and integrating content, network, and behavioural analysis significantly improves the quality of detecting coordinated information attacks. The results obtained are both scientifically novel and of practical value. They can be used to create systems for monitoring information threats, supporting cybersecurity decision-making, fact-checking, and protecting Ukraine's information space.