| 期刊全称 | Applied Computer Sciences in Engineering | | 期刊简称 | 11th Workshop on Eng | | 影响因子2023 | Juan Carlos Figueroa-García,German Hernández,Elvis | | 视频video | http://file.papertrans.cn/168/167503/167503.mp4 | | 学科分类 | Communications in Computer and Information Science | | 图书封面 |  | | 影响因子 | .The two-volume set CCIS 2222 + 2223 constitutes the proceedings of the 11th Workshop on Engineering Applications, WEA 2024, which took place in Barranquilla, Colombia, during October 23–25, 2024...The 42 full papers presented here were carefully reviewed and selected from 97 submissions. The papers are organized in the following topical sections:..Part I - Artificial Intelligence...Part II - Optimization; Simulation; Applications.. | | Pindex | Conference proceedings 2025 |
| 1 |
Front Matter |
|
|
Abstract
|
| 2 |
|
|
|
Abstract
|
| 3 |
Zero-Shot Spam Email Classification Using Pre-trained Large Language Models |
Sergio Rojas-Galeano |
|
Abstract
This paper investigates the application of pre-trained large language models (LLMs) for spam email classification using zero-shot prompting. We evaluate the performance of both open-source (Flan-T5) and proprietary LLMs (ChatGPT, GPT-4) on the well-known SpamAssassin dataset. Two classification approaches are explored: (1) truncated raw content from email subject and body, and (2) classification based on summaries generated by ChatGPT. Our empirical analysis, leveraging the entire dataset for evaluation without further training, reveals promising results. Flan-T5 achieves a 90% F1-score on the truncated content approach, while GPT-4 reaches a 95% F1-score using summaries. While these initial findings on a single dataset suggest the potential for classification pipelines of LLM-based subtasks (e.g., summarisation and classification), further validation on diverse datasets or multilingual scenarios is necessary. Besides, the high operational costs of proprietary models, coupled with the general inference costs of LLMs, could significantly hinder real-world deployment for spam filtering.
|
| 4 |
Automatic Recognition System for Public Transport Robberies Based on Deep Learning |
Laura Jalili,Josué Espejel-Cabrera,José Sergio Ruiz-Castilla,Jair Cervantes |
|
Abstract
Public transport security is a major concern in many cities, with robbery being a frequent and alarming issue. This study introduces the Automatic Alert System for Public Transport Robberies based on Deep Learning with Transfer Learning (SAATP-ATP), which focuses on analyzing audio signals. The system utilizes advanced deep learning techniques and transfer learning to enhance accuracy in detecting suspicious activities based on specific sounds associated with robberies. We evaluate the performance of various pre-trained neural network architectures on audio datasets from real robbery incidents in public transport. The system also accounts for environmental and audio quality factors to improve robustness and efficiency. Experimental results demonstrate that SAATP-ATP can identify robbery incidents with high precision, making it an effective tool for enhancing public transport security and providing rapid alerts in emergencies.
|
| 5 |
Reinforcement Learning Model Applied in a Pair Trading Strategy |
Cristian Quintero,Diego Leon,Javier Sandoval,German Hernandez |
|
Abstract
Trading strategies are in constant transformation, especially today, given the growing use of quantitative and technological bases, taking advantage of the development of methods in other areas of knowledge such as mathematics, physics or statistics, which have set a precedent either by offering greater precision or by being more efficient methods to solve problems that originated in the framework of these other branches of knowledge. It is not in vain that it is important to implement models that, given the information available in the capital markets, allow investment decisions to be made, particularly in this case in trading, and at the same time generating the highest possible profitability. However, finding a model that incorporates the greatest amount of dynamics of reality with great precision is complex, despite the fact that many advances in quantitative modeling have allowed us to get closer to this objective, especially because these dynamics are not constant and are not deterministic. It is for this reason that, in response to the need to incorporate the evolution in the market dynamics itself through the available price information, some methods such as reinforced lear
|
| 6 |
Neural Networks Informed by Physics Applied to Solving an Optimal Investment-Consumption Problem |
John Freddy Moreno Trujillo |
|
Abstract
This work considers the optimal portfolio problem of Merton in the context of an uncertain financial market composed of two types of assets. The Hamilton-Jacobi-Bellman equation associated with this type of problem is posed, and it is shown how physics-informed neural networks (PINNs) can be implemented to find its solution.
|
| 7 |
Deep Learning-Based Object Detection of Relevant Morphological Traits for Enhancing Automatic Classi |
Lilian Dayana Cruz-Cruz,Diego M. Lopez,Rubiel Vargas-Canas |
|
Abstract
Classification of freshwater macroinvertebrates is a valuable tool in environmental biomonitoring activities, although it can be time-consuming and requires specialized expertise. Macroinvertebrate samples are manually classified using taxonomic resolutions or keys to estimate the level of contamination in an ecosystem. This classification follows a hierarchy in which morphological variations in the different body parts are examined, reflecting adaptations between genotype and environment (phenotype), resulting in estimates of the quality of the resource. Automatic classification has the potential to speed up the bioindication process. However, the complexity of the morphotaxonomic traits of these organisms, the need for adequate databases to address their vast diversity, and the lack of expert confidence in using these techniques have posed significant technical challenges. Despite advances in recent years, especially with deep learning techniques, the complexity of emerging models still needs to accurately capture the fine morphological features that are key to manual taxonomic classification. This paper examines how a semantic detector like YOLO performs when dealing with fine-g
|
| 8 |
Deep Tracking Portfolios Using Autoencoders and Variational Autoencoders |
Daniel Aragón Urrego,Oscar Eduardo Reyes Nieto,Carlos Andrés Zapata Quimbayo |
|
Abstract
This paper examines the potential of deep learning techniques for the construction of tracking portfolios for the US Nasdaq 100 index. We use autoencoders (AE) and variational autoencoders (VAE) within deep neural networks, incorporating cardinality constraints to limit the number of assets in the portfolio. Our methodology entails training the models on historical stock returns from 2019 to 2022 and validating their performance in 2023. The results demonstrate that the tracking portfolios generated by both the autoencoder (AE) and variational autoencoder (VAE) models closely track the index in both the in-sample and out-of-sample periods, while reducing the number of assets. The tracking performance is evaluated in terms of cumulative returns and tracking error. Furthermore, we evaluate the performance of the AE and VAE methods across a range of asset selections, highlighting the strengths and limitations of each approach.
|
| 9 |
On the Use of a Foundation Acoustic Model to Identify Highly Relevant Phonetic Information of Parkin |
D. Escobar-Grisales,C. D. Ríos-Urrego,J. R. Orozco-Arroyave |
|
Abstract
Parkinson’s disease (PD) is a neurological condition that produces several speech deficits, typically known as hypokinetic dysarthria, affecting the production of different phonemes and resulting in an impaired speech communication. This work presents a detailed investigation based on the wav2vec 2.0 foundational model specifically tuned to perform the automatic discrimination between PD and healthy control (HC) subjects. The investigation showed that, instead of considering the complete wav2vec 2.0 architecture with 12 layers, the five layer is enough to find a model suitable to obtain good classification accuracies. Besides, this work presents a framework where frame-wise classification results are considered, enabling a detailed analysis regarding which phonemes and phonological classes are more accurate for performing the classification. All experiments are evaluated in an external and independent test set, therefore given the good results found in this work, which motivates us to continue working in this direction. For future work, we plan to modify the method to perform the time-stamp labeling to model co-articulation information in speech produced by PD patients.
|
| 10 |
Improvement in the Management of Potable Water Distribution Using Data Science for the Detection and |
Rafael Rincon Villamizar,J. L. Villa |
|
Abstract
In Colombia, an average of 41% of potable water is lost in water distribution systems, including both technical and non-technical losses. In order to reduce these losses, improving operational management is of paramount importance. This paper presents a methodology that integrates the technical expertise of operational engineers with data science to improve the management, monitoring, and maintenance of water distribution networks, focusing on efficiency and sustainability. The research introduces techniques for automated error detection and correction in large volumes of operational data, using linear regression models and deep neural networks. These models allow for the correction of missing or erroneous data in the flow measurement system, improving the precision in the assessment of the infrastructure’s condition and fostering more informed and strategic decisions. The collaborative approach between computational capabilities and expert judgment results in useful models that contribute to the optimal management of the water service and the utilization of modern technological tools.
|
| 11 |
Wrist Motion Pattern Recognition from EMG Signal Processing Using Machine Learning and Neural Networ |
Malorys M. Elles Fang,Rita Q. Fuentes-Aguilar,Y. Yuliana Rios,Duván A. Marrugo-Tobón,Sonia H. Contre |
|
Abstract
Wrist motion pattern recognition is significant in various applications, such as human-computer interaction and rehabilitation. This paper presents a study on wrist motion pattern recognition using electromyography (EMG) signal processing techniques in conjunction with machine learning. We constructed a dataset, capturing time and frequency characteristics associated with three distinct wrist movements: extension, flexion, and relaxation. The main goal of this project was to develop accurate and reliable models to classify wrist motion patterns from EMG signals. Several machine learning algorithms were used, including random forest and neural networks. We also used principal component analysis (PCA) to optimize feature selection and enhance classification performance. The results demonstrated promising outcomes for the random forest and neural network classifiers. The random forest classifier achieved an accuracy of 75%. In contrast, the neural network, specifically a multilayer neural network, achieved an accuracy of 90%. Including PCA for feature selection significantly contributed to the overall performance improvement in both classifiers. This study’s findings show the potentia
|
| 12 |
Skin Disease Pre-diagnosis with Novel Visual Transformers |
Erick García Espinosa,José-Sergio Ruiz Castilla,Farid García Lamont |
|
Abstract
There are multiple skin diseases. Skin diseases manifest through lesions. Lesions can be discolorations, rashes, or bumps. The most common ones are Smallpox, Monkeypox, Measles, Lupus, Malignant Melanomas, among others. On the other hand, the healthcare system in Mexico is structured into three levels. The first level consists of small clinics, mostly rural. These clinics have a doctor and a nurse. They only provide general consultations and vaccines. Second-level clinics offer medical assistance up to hospitalization and childbirth surgeries. Finally, third-level clinics have all specialties. A patient is evaluated at level 1 and can be referred to a level 2 or 3 clinic. This work is focused on patients who can reach a level 1 or 2 clinic. The doctor must evaluate it, but is not a dermatologist or oncologist. Therefore, we propose a mobile device called D.A.N.N (Dermatologic Analysis with Neural Networks) that aims to take a photograph of the patient’s skin and identify the disease. This is considered a pre-diagnosis. Thus, it helps the doctor decide which specialty to refer to. This device can identify up to 5 diseases. To achieve this purpose, the device was assembled with a Ras
|
| 13 |
Enhancing the Diagnostic Accuracy of Diabetes and Prediabetes with Neural Network-Based Area Under t |
Erika Severeyn,Alexandra La Cruz,Mónica Huerta,Jesús Velásquez |
|
Abstract
Diabetes is a chronic disease characterized by persistently elevated blood glucose (BG) levels, which can lead to severe complications such as heart disease, stroke, nephropathy, retinopathy, and neuropathy if left unmanaged. Prediabetes, a precursor to type 2 diabetes, is defined as a state of abnormally high BG levels that fall below the diagnostic threshold for diabetes. The oral glucose tolerance test (OGTT) is a widely used diagnostic tool for identifying individuals with diabetes and prediabetes. This test involves ingesting a glucose solution and measuring BG and insulin levels at specific intervals. Recently, researchers have begun to utilize the area under the curve of insulin (AUCI) and glucose (AUCG) of the OGTT, as diagnostic metrics. These values are calculated by measuring the area beneath the curve formed by the glucose and insulin concentration in the blood throughout the OGTT. Artificial neural networks (ANNs) have shown significant potential in enhancing the diagnosis of diabetes and prediabetes. This study explores the application of ANNs for diagnosing diabetes and prediabetes, utilizing AUCG and AUCI as diagnostic metrics. A data set of 188 individuals diagnose
|
| 14 |
Improving Energy Management in Artificial Pancreas Using an Event-Trigger MPC Strategy |
Jhon E. Goez-Mora,Pablo S. Rivadeneira |
|
Abstract
An MPC control strategy activated by events is developed, where the event determines when to perform the control calculation and, therefore, when to apply insulin to the person with type 1 diabetes. The system’s emulation environment uses the hardware-in-the-loop methodology to evaluate not only the new strategy’s performance in glycemic regulation but also the use of limited hardware resources based on a Raspberry Pi3B embedded system integrated with a battery manager and an experimental ultra-low-cost insulin infuser. The results show that the control strategy that integrates event triggering achieves regulation of test patients despite facing emulation scenarios with measurement noise, parametric variations, and without food announcements with a similar performance to the standard time-triggered strategy. The strategy permits increasing the device’s operating time up to three times compared to the controller activated in each sampling period.
|
| 15 |
Advancements in AI-Driven Emotion Recognition: A Study on CNN and DMD Methodologies |
Oscar Loyola,Diana Suarez,Griselle Salazar |
|
Abstract
In the intersection of artificial intelligence and psychology, sentiment recognition plays a pivotal role in enabling machines to comprehend human emotional expressions. Despite the advancements in artificial intelligence, accurately recognizing and responding to human emotions remains a significant challenge, particularly in diverse social environments. This research addresses this gap by comparing two innovative approaches for facial emotion classification using deep neural networks. The first approach employs a Convolutional Neural Network (CNN) to process raw images directly, while the second approach integrates Dynamic Mode Decomposition (DMD) for feature extraction prior to classification. Both models were trained and evaluated on the FER2013 dataset, and their performances were compared using confusion matrices and ROC curves. The CNN model demonstrated superior accuracy and discriminative capability, particularly in recognizing “Happy” and “Surprise” emotions, whereas the DMD model, despite lower overall accuracy, showed potential in capturing dynamic features. This study’s findings underscore the importance of combining advanced feature extraction techniques with robust ne
|
| 16 |
Control of a Buck Converter Using Artificial Neural Network NARMA-L2 Controller |
Angel Quiroga,Jhon Bayona,Helbert Espitia |
|
Abstract
This document studies the aspects of using a NARMA-L2 controller based on neural networks for the regulation of a buck converter widely used in energy transformation. The implementation is carried out at the simulation level in MATLAB using Simulink where the design and the results are described, observing that this type of controller can be used in energy conversion applications.
|
| 17 |
Design of a Crime Prediction Model for Barranquilla Using Machine Learning Algorithms |
Rodríguez Morales Jeison Estiven,Estupiñan Gomez John Fredy,Carreño Hernandez Pablo Enrique,Simanca |
|
Abstract
One of the great concerns in the country is the high frequency of crime in urban areas; Barranquilla is among the 5 most populated cities in Colombia where a variety of crimes affect the population, generating in them a growing perception of insecurity over the years. This article presents a model according to the CRISP-DM methodology, which seeks to analyze the crime in Barranquilla and predict their behavior at different times of the year and in different areas of the city. The Dataset comes from the Metropolitan Police of Barranquilla and contains 31763 records from 2018 to 2022. The results obtained from this research will allow the Metropolitan Police to make decisions for the safety of citizens.
|
| 18 |
A Smart Mobile Mapping Application for the Evaluation of Road Infrastructure in Urban and Rural Corr |
Diego Espinel-Gomez,Wilmar Fernandez-Gomez,Julian Moreno-Moreno,Daniel Carranza-Leguizamo,Camilo Mar |
|
Abstract
Mobile mapping systems are widely used for urban observation. By integrating electronic devices with various sensors (including optical and thermal), computer vision techniques, AI algorithms, and GNSS positioning systems, data from urban road corridors can be collected. This research presents a review of conventional mobile mapping methods and proposes an advanced computer vision approach that leverages AI to improve mapping accuracy. The limitations and challenges of traditional methods, such as vehicle-mounted LIDAR systems, are critically examined. The study highlights how AI integration, particularly deep learning algorithms, has significantly enhanced data interpretation and decision-making processes in road infrastructure management. An experiment was conducted on Bogotá‘s road network using a mobile mapping system equipped with video sensors and GNSS devices in a specially adapted vehicle. This setup facilitated the collection of large datasets, which were subsequently processed using the YOLO AI algorithm to effectively detect and classify road pavement conditions. The experiment‘s results underscore the effectiveness of combining mobile mapping technology, programming, an
|
| 19 |
Analysis of Variables Related to Criminal Violence and Public Insecurity in the City of Barranquilla |
C. William R. Insignares,B. Emeldo Caballero,I. Leidy M. Mora,H. Pablo E. Carreño,Aldo Silvera |
|
Abstract
This study examines the behavior of crime rates in Barranquilla over the period 2018 to 2022, with a specific focus on temporal variations and according to the type of crime. To conduct this research, official data from government sources were used and subjected to statistical analysis in order to identify patterns and trends..During this period, the Colombian city of Barranquilla has experienced a significant increase in its crime rate, particularly with regard to robberies and robberies with violence, which has generated a growing concern for the safety of its inhabitants. This increase in crime is attributed to several factors, including poverty and lack of economic opportunities, which increase the risk of criminal activity. In addition, the presence of well-structured criminal groups has also contributed to higher crime rates in the region. Despite efforts by law enforcement authorities to combat these groups, they face difficulties due to limited resources and the presence of corruption at some levels..The methodology used in the research is characterized by being analytical and of mixed approach, where a four-phase process was carried out: descriptive, data cleaning, interve
|
| 20 |
Comparison of Motor Imagery and Motor Execution Networks Using the Phase Lag Index |
Mateo Alzate-Márquez,Andrés Quintero-Zea |
|
Abstract
Electroencephalography has long served as a pivotal modality in the exploration of brain functions, particularly within the domains of motor imagery and motor execution. This investigation aimed to delineate commonalities and disparities in functional connectivity networks of motor imagery/execution by employing the weighted phase lag index for the analysis of EEG data across the mu and beta frequency bands. The results reveal that the topological configurations of brain networks at the functional level exhibit consistent characteristics, particularly with respect to leaf fraction, diameter, and maximum degree, in both motor imagery and execution tasks. In general, the findings highlight the critical need to incorporate multiple features and frequency bands in the evaluation of topological metrics within the realms of motor imagery and execution. These insights are poised to substantially advance our understanding of the neurological foundations of these motor activities and contribute to future scholarly inquiries and technological advancements in neurotechnology.
|
|
|
|Archiver|手机版|小黑屋|
派博传思国际
( 京公网安备110108008328)
GMT+8, 2025-9-25 21:59
发表于 2025-3-21 16:09:56
