Energy consumption planning under environmental and economic uncertainty has become a strategic challenge, particularly after disruptive events such as the COVID-19 pandemic. This study develops and compares conventional polynomial regression and deep learning architectures to simulate and forecast national energy consumption in Mexico using long-term data from 1965 to 2021. The proposed framework integrates environmental (CO2 emissions, temperature increase), economic (GDP per capita), and energy infrastructure variables (number of power plants, oil production). For simulation purposes, single-layer and multilayer Artificial Neural Networks (ANN and MLFFNN) achieved the highest accuracy with R 2 = 0.9999 and MAPE values below 0.38 %. For forecasting, polynomial regression outperformed the RNN-LSTM model with R 2 = 0.9339, whereas the LSTM architecture showed better adaptability in capturing abrupt consumption drops during the pandemic period. The comparative evaluation using R 2, RMSE, MAPE and paired t-tests demonstrates that combining statistical and deep learning approaches improves robustness in long-term forecasting scenarios. The proposed multi-approach framework provides a data-driven decision-support tool for intelligent energy planning under emergency conditions.
Over time and technological advancements, and across all levels, some words disappear while others are added to the lexicon. This vocabulary, both old and new, has become a unified set that humans can use within their surrounding environment simultaneously. This emergence or disappearance has led to the addition of some words that differ radically in pronunciation and spelling, yet simultaneously carry a single or similar meaning, a phenomenon known as synonymy. On the other hand, some words, through time and technological development, have more than one meaning when written, a phenomenon known as homonymy. This difference and variation between the two fields has led to confusion in linguistic information when used. This research paper addresses some of these words using information technology within information networks and specialized databases. These words have been added, integrated, and archived into a specialized linguistic dictionary, thus unifying their intended meaning within the context in which they are used for this purpose. Healthcare IT is an important component of data management, analysis, and exchange within the distributed network of healthcare services. The problem is that terminology discrepancies, which include both synonyms and homonyms, are still a stumbling block for reaching semantic interoperability. The proposed approach in this study is an ontology framework that is a combination of the two. This framework will be specifically applied in Healthcare IT and is intended to solve terminology problems before undertaking semantic data integration. The suggested solution not only facilitates data processing but also ensures interoperability and structured knowledge management within healthcare systems. It is by eliminating terminological discrepancies during the initial phases that this theoretical model helps ensure reliable data exchange and intelligent information processing in the healthcare IT infrastructure.
This study evaluates the performance hierarchy of AI chatbot services in replacing customer service representatives using the analytic hierarchy process (AHP). Five criteria cost efficiency, accuracy, speed, scalability, and human impact – were analyzed across three alternatives: AI Chatbots, voice recognition systems, and hybrid models. Results show that the hybrid model achieved the highest overall score (0.823), followed by AI Chatbots (0.81) and voice recognition systems (0.766). AI Chatbots excelled in scalability and speed, while the hybrid model performed best in accuracy and human impact, demonstrating its ability to balance automation with ethical considerations. The findings highlight that integrating human oversight within automated systems enhances service adaptability, reliability, and user satisfaction. This research contributes to understanding how hybrid AI frameworks can optimize performance while maintaining empathy and trust key elements in the transition toward AI-driven customer service.
This study presents a machine learning approach for identifying encryption algorithms based solely on ciphertext analysis. The problem is formulated as a supervised classification task, where features are learned directly from ciphertext samples generated by four widely used symmetric encryption algorithms. Experimental results show that the model can reliably distinguish between ciphers under constrained conditions, particularly when initialization vectors (IVs) and key distributions remain consistent across samples. These findings indicate that ciphertexts retain subtle statistical patterns that can be exploited for algorithm identification, while also highlighting limitations in generalization under distribution shifts.
Precise prediction of drug–drug interactions (DDIs) is crucial to maximize therapeutic efficiency and ensure patient safety, particularly in polypharmacy, where the use of multiple drugs is common. This paper proposes a new hybrid deep learning (DL) model for multiclass DDI classification, structured in two stages. In the first stage, a dual-branch convolutional neural network (CNN) with shared weights, inspired by Siamese architecture, is used to extract pairwise similarity-based features by utilizing Jaccard similarity vectors computed from protein targets, enzymes, and molecular structure fingerprints derived from the Simplified Molecular Input Line Entry System (SMILES). Simultaneously, the graph neural network (GNN) encodes molecular graph representations through the same SMILES. In the second stage, these complementary feature paths are combined into a single embedding and input into a fully connected classifier. The model has been improved through residual connections to allow easier gradient flow, a focal loss to solve class imbalance, and a composite Rectified Adam (RAdam)- LookAhead optimizer to enhance convergence and training stability. Empirical results conducted on a dataset comprising 572 drugs and 37,265 drug interactions in 65 DDIs illustrate that the proposed model achieves significantly superior performance to state-of-the-art baseline methods, with an accuracy of 96.28 % and an F1-score of 96.28 %. The findings confirm that the combination of structured similarity-based features and molecular topology features yields a scalable solution for high-accuracy DDI prediction, making a significant contribution to the development of pharmacovigilance tools.
