The problem of poor air quality in the Amazon is a serious issue, as air pollution in the region negatively affects public health, resulting in thousands of premature deaths and severe damage to the environment. Monitoring emissions is crucial for enforcing laws that restrict these emissions and for preventing fires and their devastating consequences. For this reason, an air quality monitoring network has been implemented in the Amazon region, currently with several sensors distributed throughout the state of Acre/Brazil. However, many sensors have significant data gaps, in some cases with more than 80% loss. This is due to power failures, internet connection problems and device defects, thus compromising the consistency and accuracy of air quality measurements. This paper investigates the use of imputation techniques applied to estimate missing data from Amazon sensors collected from January 1, 2020 to December 31, 2023. Simple imputation techniques (Mean, Median) and those based on machine learning (MICE, KNN and MissForest) were selected. In the experiments, missing data was randomly introduced into the complete dataset (from 10% to 50%), and the techniques were compared using the following evaluation metrics: Mean Square Error (MSE), Root Mean Square Error (RMSE) and coefficient of determination (R²). The results showed that advanced techniques such as KNN and MICE are superior to simpler techniques, with lower MSE and RMSE, as well as a higher R². Even for the most critical case (50% missing data), KNN achieved an MSE of 0.0013 and an R² of 0.85, and MICE presented an MSE of 0.0013 and an R² of 0.93, standing out as effective methods for data imputation.

In this study, we leveraged a publicly available dataset containing 3256 peripheral blood smear (PBS) images, prepared in the bone marrow laboratory of Taleqani Hospital in Tehran, Iran. This dataset consists of blood samples from 89 patients suspected of Acute Lymphoblastic Leukemia (ALL). The images were captured using a Zeiss camera at 100x magnification and stored as JPG files. The dataset is divided into two primary classes: benign hematogones and malignant lymphoblasts. The malignant lymphoblasts are further categorized into three subtypes: Early Pre-B, Pre-B, and Pro-B ALL. The definitive classification of these cell types and subtypes was performed by a specialist using flow cytometry tools.

To classify these images into four distinct categories, we employed a stacked ensemble learning approach. Our model stack included three base models, DenseNet121, VGG16, and VGG19, with a K-Nearest Neighbors (KNN) classifier acting as the meta-model. This ensemble method capitalizes on the strengths of each individual model to improve overall classification performance. Our approach achieved a high accuracy of 94%, demonstrating its robustness and reliability in distinguishing between the various cell types and subtypes within the dataset.

The significant accuracy attained underscores the potential of advanced machine learning techniques in medical image analysis, particularly in the context of hematological malignancies. Our findings suggest that such methodologies could greatly enhance diagnostic precision and efficiency, leading to better patient outcomes. This study illustrates the promising application of deep learning models in the automated classification of ALL subtypes, paving the way for future advancements in the field.

Controlling infectious diseases in animals on farming holdings plays an important role in ensuring livestock health and productivity. Infectious diseases can spread rapidly among animals, leading to severe consequences such as reduced productivity, increased mortality, and substantial economic losses. Therefore, implementing effective disease control measures is crucial for safeguarding animal welfare and farmers' livelihoods. When animals are kept outdoors in large areas, identifying diseases and their sources of contamination can be particularly challenging. The vastness of these environments makes it difficult to monitor every animal closely and detect early signs of illness. Additionally, the mingling of animals from different areas can facilitate the spread of diseases, making it harder to pinpoint and control outbreaks. This paper presents an architecture designed to mitigate these challenges. To sum up, this solution uses a set of IoT sensors that try to identify the proximity of healthy animals to animals with disease to stop transmission. The IoT sensors will provide farmers with real-time data, enabling them to swiftly isolate infected animals and implement targeted interventions. By improving disease detection and monitoring, this technology will help farmers maintain healthier herds and reduce the risk of widespread outbreaks, thereby enhancing both animal welfare and agricultural productivity significantly.

Software package management tools have become common and are available for practically all SDKs. They allow for the definition of dependencies between packages, ensuring consistent use of their respective versions, especially during installation, updating, configuration, and removal. These tools are primarily used in the software development phase by programmers. While the utility of software package managers and the added value they provide to programmers during the development stage are unquestionable, there are still many gaps concerning the remaining phases of the software lifecycle—commonly referred to as the maintenance stage. The need for maintenance arises from the outdatedness of packages, resulting from incompatibilities with other packages, the introduction of improvements and optimizations, the correction of errors, the elimination of vulnerabilities, and so on. Although it is usually possible to identify packages that are deprecated or obsolete, updating is still a manual process initiated by the programmer. In this paper, authors propose a solution, still in its prototype stage, aimed at equipping applications with the means to report their status concerning update needs, particularly for critical updates. The solution consists of a background service that processes technical reports published by various sources, an ontology used to standardize information and concepts from responsibility disclosure reports, a REST service used by applications to obtain a self-diagnosis of their condition and a REST client that is automatically installed in the application.

Smart cities are made to provide services, including software services, for citizens.. There are many services that these cities provide, and the number of active users and connected devices may vary. Traditional approaches to software design and development do not take into account both the high-level and low-level management of complex services, which include IoT devices, real-time applications, and AI-related processing frameworks. Some of the most important components of software system management are internal task routing and services management. Smart city systems are developed and implemented for real-world cities using real-world data. For example, Kyiv city is the largest city in Ukraine in terms of population and total area. In this study, data on Kyiv city are used as a foundation of the proposed smart city system model. We provide a generalized software service model that is based on control parameters, benchmarks, and weights. These services are provided in an algorithmic and software model version that allows for their implementation in already existing services or when developing new smart city solutions. The smart city platform is based on the integration of various components; in turn, each individual component consists of its own set of software and hardware services and components. The tasks of sub-services and module management are as crucial as they are complex. The system manager module is a middleware/core-layer software system. Event handling, routing, and service/process activation are determined by the appropriate mathematical calculation mechanism. This role can be filled by special routing and management services, each being platform- and deployment-agonistic. While routing services are designed following standard protocols, APIs, and middleware-layer services, sub-service management systems are low-layer data/process processing- and computation-first systems.

Managing user comfort preferences in a smart environment presents unique challenges due to conflicting requirements and expectations. This paper explores innovative strategies to harmonize diverse user preferences within shared smart spaces. As smart environments become increasingly prevalent in homes, offices, and public buildings, the need to accommodate individual comfort settings for temperature, lighting, and noise while minimizing conflicts among users becomes critical.

This study investigates a specific case within a multi-occupant smart building, analyzing how conflicts in comfort preferences are identified, addressed, and resolved. By implementing a dynamic preference management system, which utilizes machine learning algorithms and real-time data analytics, the proposed solution aims to balance and optimize individual comfort levels. The system considers historical data, context-aware adjustments, and predictive modeling to preemptively address potential conflicts.

The findings demonstrate that integrating advanced computational techniques with user feedback mechanisms significantly enhances the overall comfort experience. The research highlights the importance of adaptive systems that can learn and evolve with user preferences, ultimately leading to more harmonious coexistence in shared smart environments.

This paper contributes to the field of smart environment management by providing a comprehensive framework for conflict resolution and offering practical insights into the deployment of user-centric comfort management systems. The case study underscores the potential of technology to create more responsive and personalized smart environments that cater to the diverse needs of their occupants.

The video projector is an essential tool for various activities such as teaching, organizational tasks, and conferences. This article introduces a low-cost and effective architecture designed to enhance video projection resources, including older models, through the incorporation of virtualization for improved management and sharing capabilities. The proposed solution addresses prevalent connectivity issues caused by differing connector types, transmission protocols, and configuration incompatibilities (such as frequency and resolution) between video projectors and computers. These incompatibilities frequently lead to delays and challenges in effectively utilizing video projection resources.

The innovative architecture utilizes a Raspberry Pi combined with three virtualized applications to create a user-friendly system. This system not only facilitates the efficient management of video projection resources but also allows for seamless sharing and connectivity across various devices. By leveraging virtualization, the architecture ensures compatibility and adaptability, reducing the downtime typically associated with setup and configuration issues.

The implementation of this solution is aimed at enhancing the functionality and accessibility of video projectors, enabling new paradigms in working and teaching environments. The approach provides a low-cost and practical method to upgrade existing video projection infrastructure, thereby extending the lifespan and utility of older projectors while introducing modern capabilities and improving overall user experience.

New technologies solve complex problems in agricultural production, enabling both food security and the optimization of the use of natural resources. In this paper, we will cover hydroponic agriculture, advanced monitoring, and data analysis. Plants interact with light, humidity, and temperature environmental factors. Light is one of the most important environmental factors for photosynthesis and growth. Various wavelengths have contrasting impacts on plant development and physiological traits. Artificial manipulation of light conditions may allow for an improvement in crop production and quality. Thus, the present study deals with the influence of diverse light wavelengths on the growth of strawberry plants cultivated in a hydroponic system. The growing processes of strawberry plants (Fragaria × ananassa) have been taken into consideration due to their importance as food and the sensitivity of their cultivation. It tests the influence of specific wavelengths on plant growth and development. State-of-the-art technologies include Arduino-based temperature and light sensors to monitor the cultivation conditions in real-time, while Convolutional Neural Networks trace the growth patterns and pests of the crops by images taken. This work models and predicts behaviors of plants under different light conditions using Machine Learning techniques, thus optimizing cultivar development with a view of maximum yield production. The results obtained show that red light promotes growth through enhanced flower and fruit development. Blue light is favored by robust leaf and stem growth since it is most effective in photosynthesis. Green lights, which help inner light penetration inside the leaf canopy, have less of an effect on photosynthesis. Yellow light also has some advantages in general growth but is inefficient compared with blue and red light. Result using CNN architecture, accuracy 89%. This work contributes to precision agriculture, sensor technology, and sustainable farming practices.

Regulatory peptides are widely used as immunomodulators, neuroprotectors, and neurometabolicstimulators. This work is devoted to the isolation of peptide complexes from the epidermal secretion of ascaleless fish species and the study of these complexes properties. It has been suggested that suchpeptides can suppress inflammation and promote the regeneration of human integumentary tissue.The skin secretion of the African catfish (Clarias gariepinus) was taken as a source of low-molecularpeptides. Peptide isolation was carried out according to the following scheme: acetic acid extraction inthe presence of a biogenic salt (MgCl2, CaCl2, ZnCl2, MgSO4, CaSO4); ultrasonic cavitation treatment;removal of major proteins by filtration; precipitation of low-molecular weight peptides using acetone.To qualitatively check the resulting product, IR spectroscopy and chromatomass spectrometry wereused. Using these methods, it was possible to prove that the isolated product is a complex containingtens of thousands of low molecular weight peptides. To confirm the antioxidant properties of theproduct, the autoxidation of adrenaline in an alkaline medium was used. Antioxidant properties aredecisive in wound healing preparations of peptide origin. It was found that the resulting peptidepreparation is capable of inhibiting the oxidation reaction of a 1% solution of adrenaline in a buffersolution, which clearly indicates its antioxidant properties. It has been shown that the use of CaCl2makes it possible to isolate the peptide fraction with the highest antioxidant activity.

In the era of big data, the sheer volume of information available for machine learning applications has grown exponentially. However, this increase in data often leads to a decrease in the quality of datasets due to issues such as noise, redundancy, and irrelevance, which can adversely affect the performance of predictive models. To address these challenges, dimensionality reduction techniques are employed, with feature selection being a prominent method. The objective of this research is to enhance machine learning prediction performance by developing a novel hybrid feature selection algorithm. This algorithm synergizes the strengths of cat swarm optimization (CSO) with those of the crow search algorithm (CSA), aiming to refine feature selection processes. The proposed hybrid algorithm was meticulously implemented and applied using the K-Nearest Neighbors (KNN) model on a diverse array of datasets. To rigorously evaluate its efficacy, the algorithm was tested on 12 carefully selected datasets, encompassing various domains and complexities. The performance was measured based on the accuracy of the machine learning predictions. Remarkably, the proposed hybrid algorithm achieved an average accuracy rate of 87%, which represents a significant improvement over previous approaches that had an average accuracy of 83%. These results underscore the potential of the proposed hybrid feature selection algorithm in enhancing the predictive capabilities of machine learning models by effectively reducing dimensionality and eliminating irrelevant features. The findings suggest that integrating CSO and CSA can lead to more robust feature selection mechanisms, thereby improving the overall quality and reliability of machine learning predictions.