Water quality is essential to the population’s well-being, water resources management, and environmental development strategies. In this article, we propose a framework based on machine learning (ML) techniques for enhancing the assessment of water quality based on water quality indices (WQIs). It consists of three algorithms that could serve as a foundation for automating the evaluation of any resource based on indices and can operate locally or globally. Local-level algorithms assist in selecting suitable WQIs tailored to specific water sources and quality requirements, while global-level algorithm evaluates WQI robustness across diverse water sources. We also provide a warning system to mitigate differences in water quality evaluation using WQIs and a valuable tool (based on the features’ importance) for selecting ML models that prioritize the water parameters’ significance. The framework’s design draws upon conclusions from a case study involving the forecast and comparison of two WQIs for the Brahmaputra River. Any other data series, WQIs, and water parameters can be employed.

Prediction of the quality of the result provided by a specific solving method is an important factor when choosing how to solve a given problem. The more accurate the prediction, the more appropriate the decision on what to choose when several solving applications are available. In this article, we study the impact of the structure of a Traveling Salesman Problem instance on the quality of the solution when using two representative heuristics: the population-based Ant Colony Optimization (ACO) and the local search Lin-Kernighan (LK) algorithm. The quality of the result for a solving method is measured by the computation accuracy, which is expressed using the percent error between its solution and the optimum one. We classify the instances in structured, semi-structured, and unstructured and perform a between classes and inside-classes analysis. All the structured instances were solved to optimality by the ACO implementation, which was not the case for the LK application. On small random instances, the ACO implementation used in this paper also optimally found the solutions. We show that the quality of the results on semi-structured and unstructured instances can be predicted using some instance parameters when using the ACO implementation. Using the same parameters, the accuracy of the solutions provided by the Lin-Kernighan application cannot be predicted. We also propose several new structured, clustered, and unstructured 2D Euclidean Traveling Salesman Problem instances that can be used by the research community for further investigations.

Computing systems grouped in subnets use distributed security models, in general, by creating session keys based on the Diffie–Hellman model, and calculating the necessary parameters for this, on each of the systems. In the particular case of a network of devices heterogeneous in terms of computing power, such as IoT, the modeling of a security system of the entire structure will have to take into account the fact that some devices have a very low computing power. In this sense, starting from the study of some general models, used in structures of this type, an integrated structure was developed to secure communications and test certain vulnerable components, to calculate a degree of risk that they are maliciously intended. The system was developed with a customized mathematical model, a scheme for propagation and management of cryptographic parameters and a test in a real environment by creating the algorithmic model and implementing it within a structure of a beneficiary.

In the digital era, the need for precise and efficient search operations is paramount as users increasingly rely on online resources to access specific information. However, search accuracy is often hindered by errors in user queries, such as incomplete or degraded input. Errors in search queries can reduce both the precision and speed of search results, making error correction a key factor in enhancing the user experience. This paper addresses the challenge of improving search performance through query error correction. We propose a novel methodology and architecture aimed at optimizing search results across thematic websites, such as those for universities, hospitals, or tourism agencies. The proposed solution leverages an intelligent model based on Gated Recurrent Units (GRUs) and Bahdanau Attention mechanisms to reconstruct erroneous or incomplete text in search queries. To validate our approach, we embedded the model in a prototype website consolidating data from multiple universities, demonstrating significant improvements in search accuracy and efficiency.