International Conference on Applied Technologies

Malaria is a disease caused by parasites of the genus Plasmodium, which can be transmitted through the bite of a mosquito or even through transfusions of infected blood. This represents one of the main health problems for developing countries. Despite the existence of conventional procedures for the eradication and control of this disease, there are still some challenges in geographically remote areas, where the lack of accessibility, equipment and specialized training can lead to inadequate treatments. From this perspective, early detection of this illness is crucial. Faced with these challenges, recent artificial intelligence advances offer new opportunities to optimize the identification of diseases and pathologies. Along these lines, the aim of this paper is to develop a tool based on convolutional neural network to support the detection of malaria using blood samples. In this context, a set of images was identified from the Kaggle platform. This set identifies two types of conditions for blood samples: malaria infection, non-malarial infection. Then, through CRISP-DM (Cross Industry Standard Process for Data Mining), this research evaluated four convolutional neural networks and chose the most optimal one to implement within a web application. In this sense, VGG-16 stood out for achieving an accuracy of 94%, while Inception V3, RestNet-50, and InceptionResNet obtained 91%, 58%, and 92% respectively. Conclusions and future work are detailed at the end of the paper.

Determining a reliable model to predict the compressive strength of a geopolymer is important for finding better options for Portland cement, which has many disadvantages such as weaker physicochemical properties and a production with a large carbon footprint. For this purpose, an Exploratory Data Analysis (EDA) was performed with the information from experimental tests. This EDA allowed training an accurate machine learning model for compressive strength prediction. Various individual models and mixtures of models were tested, and the decision was made to utilize the best-performing individual model. The obtained results are similar to the literature in which the best models are based on decision trees and boosting algorithms. In addition, a simple and user-friendly interface was developed for making predictions using the selected model.

Bullying, or school bullying, is a severe problem that affects the emotional well-being and social development of students, manifesting itself in forms such as physical aggression, verbal abuse, and cyberbullying. Early detection of this behavior is challenging, as it often goes unnoticed by human observers. This study proposes using LLaMA2, a generative artificial intelligence model, to identify early signs of bullying in the verbal interactions of seventh-grade students. By analyzing large volumes of data, LLaMA2 can detect signs of bullying that humans might miss. LLaMA2 will analyze conversations and provide detailed information about the dynamics of bullying, using comparative evaluations of data in statistical graphs. This approach reveals the prevalence and impact of bullying. The findings show specific patterns of bullying, combining quantitative and qualitative analyses for a deeper understanding of the problem. The results highlight that integrating artificial intelligence with traditional methods can improve the detection and management of bullying. The use of LLaMA2, together with statistical analysis, facilitates the design of personalized interventions and the identification of emerging patterns. Cooperation between educators and technology developers is crucial to maximizing impact and creating a safe and healthy environment. Continuous feedback between users and the AI system allows for ongoing adjustments. This comprehensive approach improves bullying management and fosters an inclusive and safe learning environment.

A 2023 TomTom study identifies Lima as the most congested city in Latin America and the fifth worldwide, causing significant impacts such as increased stress, long travel times, and lower productivity. Despite various approaches, none fully address the unique challenges of developing cities like Lima. To address this, the proposed system uses the Max Pressure (MP) algorithm for traffic signal control and a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network for traffic prediction. The MP algorithm dynamically adjusts signal timing to optimize traffic flow, while the Bi-LSTM predicts future traffic patterns. Validated through simulations on Javier Prado Avenue, the system achieved a 13.8% increase in vehicle throughput and an 18% reduction in travel times. These findings highlight its potential to enhance urban traffic management. The developed web app offers a practical tool, bridging research and real-world application to address Lima's traffic congestion effectively.

Alzheimer's disease (AD) is an advanced neurodegenerative disorder designate by cognitive decline and memory loss, posing significant challenges for early diagnosis and intervention. Recently, Artificial Intelligence (AI) techniques have emerged as promising tools for AD prediction. In this paper, we compare two prominent learning models i.e. Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) for AD prediction. CNN model leverages its ability to inevitably learn hierarchical features from raw imaging data, employing multiple convolutional and pooling layers to extract spatial patterns indicative of AD progression. Meanwhile, the SVM model utilizes a linear or polynomial kernel to construct an optimal decision boundary between AD-positive and AD-negative cases based on handcrafted features derived from the imaging data. When evaluated, the CNN model attained an accuracy of 87.41% with a precision of 88.41%, recall of 87.41%, and F1 score of 87.57%, whereas the SVM model demonstrated outstanding performance, with a training accuracy of 100% and testing accuracy of 98.83%, precision of 99.21%, recall of 98.44%, and an F1 score of 98.83%.

This initiative uses tailored feedback systems and one-on-one talks to inspire pupils. We use reinforcement learning to compare the approach to conventional e-learning platforms, individualized learning systems, and online classrooms. We examined success elements such as user enjoyment, interest rate, learning outcomes, and emotional connection. The recommended strategy frequently outperforms well-known ones in key student achievement areas. A user satisfaction score of 92, an engagement rate of 88, and learning outcomes of 85 demonstrate the effectiveness of tailored interactions. You may construct a flexible and helpful learning environment with high interaction quality and task completion ratings. This research indicates how crucial it is to employ innovative teaching approaches, bridge gaps, and create effective learning systems that satisfy all students’ requirements. Last, it teaches us how to alter teaching strategies to help all students succeed.

In this research, we propose real-time health tracking in IoT-based healthcare systems. Continuously gathering, processing, and interpreting biological data from different sensors is the goal. The approach focuses on signal preparation, including noise removal, normalization, and frequency domain translation for further analysis. Advanced methods include statistical thresholding, harmonic analysis, and signal quality monitoring, which may identify health issues like rhythms. The technique offers outstanding accuracy, sensor sensitivity, and minimal battery utilization, improving real-time smart health monitoring systems. A complete research study reveals that the recommended technique outperforms others in reaction time, signal-to-noise ratio, sensor durability, and real-time data processing. This strategy keeps patient data sensitive and secure. High comfort, extended sensor lifespan, and cost-effectiveness make the proposed technique a solid performance-price balance. This makes it suitable for most. It improves IoT-based healthcare systems, provides precise health monitoring, and speeds up medical treatments. When coupled, these technologies enable tailored healthcare advancements that will enhance patient outcomes via consistent, effective, and safe health monitoring.

This research uses deep learning and emotion analysis to detect social media hate speech better. The outcome is a reliable, widely usable system. The first step involves editing raw social media content. Machine learning models employ feature vectors derived using TF-IDF. The algorithm predicts sentiment scores and adds them to create a sentiment profile for grouping. Tuning hyperparameters improve model performance and detect harmful compounds. A weighted feature vector machine learning model improves sentiment-based classification. A loss function and gradient descent improves the model. Deep learning models boost identification accuracy, helping the system manage complicated real-time social media data. The proposed method outperforms NLP, supervised machine learning, and rule-based systems in accuracy, precision, and memory. The system can manage massive volumes of social media data in real time and is difficult to hack as user numbers expand. You can trust Indian internet laws and safety. This technology helps us identify bogus news and hate speech, improving the internet while adhering to the law and morality.

A systematic literature review, enhanced by scientific mapping of available literature in the Scopus database, was conducted using the Bibliometrix module in Rstudio software and VOS Viewer software, whilst adhering to PRISMA publication guidelines. This review focused on the design of a repository and Geovisor for disseminating data, products and information related to research on the biophysical assessment of land degradation dynamics and recovery practices. The review was refined through the application of Eric thesaurus terms or equivalent terminology and Boolean operators. The research variables encompassed scientific output, authors, information sources, collaboration networks, bibliometric indicators, and sponsoring institutions and countries, amongst others. The findings reveal a geographical concentration and imbalance in scientific production, whilst the collaboration network, though complex and fragmented, is predominantly composed of researchers from the United States, United Kingdom and China. The responsible conduct of researchers in integrating metadata into their scientific outputs enables the effective use of bibliometric analysis and exploitation tools. Looking ahead, there is a recognised need to develop studies employing alternative tools for generating scientific maps within the specified context.

The lack of effective adaptability in computer systems when dealing with people with motor disabilities hinders productivity and educational growth. Through this study, a review of articles on the development of Human-Computer Interaction (HMI) is conducted, arriving at studies involving the use of eye-tracking technologies as an advance for HMI. Consequently, we seek to design a prototype eye-tracking software that allows these individuals to interact with computers using their visual capabilities and evaluate the prototype to analyze future improvements. Various eye-tracking technologies are explored to identify the most effective one, considering accuracy, response times, and ease of implementation. The prototype prioritizes simple and effective control, ensuring intuitive and fluid interaction. As a methodology, initial tests of the prototype are performed with users with motor disabilities to evaluate the usability and workload of the software. SUS and NASA-TLX questionnaires collect quantitative and qualitative data on user experience. The results indicate that the prototype is marginally acceptable in terms of usability, although it presents a high workload due to mental and temporal demands. Despite these challenges, eye tracking shows a promising future as an assistive technology for people with motor disabilities. The research seeks to advance technological inclusion and promote equal opportunities for people with motor disabilities, facilitating their participation in digital activities and improving their quality of life. It is expected that the eye-tracking software developed will contribute to bridging the digital divide and promote a more accessible society.

Ankle fractures, common in high-risk sports, cause significant pain and can be treated with mobilizers. This study presents the development and validation of a continuous passive motion prototype for ankle rehabilitation. Based on mechatronic design methodology, the prototype focuses on dorsiflexion and plantar flexion movements. The design includes foot support and a gliding system inspired by backhoe excavator mechanisms. Computer-aided design, material selection, and finite element analysis in Solidworks were used to conceptualize, visualize, and simulate the prototype. Safety factors of critical system components exceeded safety limits to ensure structural integrity. Digital PID control with an RST controller achieved precise angular position control, complemented by a motor speed control algorithm for fast, precise motion. The prototype, constructed with appropriate materials, accommodates different foot sizes and provides comfort and safety during therapy. Professional physical therapists and end-users (without clinical history) performed quantitative validation. The evaluation focused on physical adaptation, functional range of motion, hygiene, protocol compliance, feasibility, ergonomics, reliability, design, and perceived safety. Structured surveys indicated good acceptance, with average ratings ranging from 3.5 to 4.8. A positive perception of the surveyed population was obtained, which implicates the prototype’s feasibility in therapeutic applications. The study highlights the potential of mechatronic design in developing effective rehabilitation devices and offers a promising solution for ankle fracture recovery in sports medicine and physical therapy.

Fires are a challenge that every country encounters, with consequences that can be either minimal or devastating. In high-rise buildings, responding to fire emergencies becomes much more complex, as there may not be enough quick and efficient response tools available. In recent years, rapid urban development in South American countries has heightened the risk of severe outcomes from urban fires, as not all nations are fully equipped to handle such emergencies. The objective of this study is to explore the potential of drones in addressing high-rise fire emergencies, with a specific focus on people detection algorithms and fire mitigation mechanisms. A systematic literature review was carried out on firefighting drones, with the guidance of five research questions and the use of PRISMA and PICo methodologies.