ABSTRACT Wind energy is an important renewable source that has been considerably developed recently. In order to obtain successful 24-h lead-time wind power forecasts for operational and commercial uses, a combination of physical and statistical models is desirable. In this paper, a hybrid methodology that employs a numerical weather prediction model (Weather Research and Forecasting) and a neural network (NN) algorithm is proposed and assessed. The methodology is applied to a wind farm in northwestern Mexico, a region with high wind potential where complex geography adds large uncertainty to wind energy forecasts. The energy forecasts are then evaluated against actual on-site power generation over one year and compared with two reference models: decision trees (DT) and support vector regression (SVR). The proposed method exhibits a better performance with respect to the reference methods, showing an hourly normalized mean absolute percentage error of 6.97%, which represents 6 and 13 percentage points less error in wind power forecasts than with DT and SVR methods, respectively. Under strong synoptic forcing, the NN wind power forecast is not very accurate, and novel approaches such as hierarchical algorithms should be employed instead. Overall, the proposed model is capable of producing high-quality wind power forecasts for most weather conditions prevailing in this region and demonstrates a good performance with respect to similar models for medium-term wind power forecasts.

RESUMEN La energía eólica es una importante fuente de energía renovable que se ha desarrollado considerablemente en los últimos años. Con el fin de obtener pronósticos de energía eólica de alta precisión con 24 horas de anticipación para usos operativos y comerciales, se necesita una combinación de modelos físicos y estadísticos. En este artículo se propone y evalúa una metodología híbrida que emplea un modelo numérico (Weather Research and Forecasting) y un algoritmo de red neuronal (NN). La metodología se aplica a un parque eólico en el noroeste de México, una región con alto potencial eólico donde la compleja geografía agrega gran incertidumbre a los pronósticos de energía eólica. Los pronósticos de energía se evalúan usando la generación de energía real en el sitio de interés durante un año y se comparan con dos modelos de referencia: árboles de decisión (DT, por su sigla en inglés) y regresión de vectores de soporte (SVR). El método propuesto muestra un mejor desempeño respecto a los métodos de referencia, mostrando un error absoluto medio porcentual horario de 6.97%, lo que representa 6 y 13 puntos porcentuales menos de error en los pronósticos de energía eólica que con los métodos DT y SVR, respectivamente. El pronóstico de energía eólica NN no es muy preciso cuando tiene un fuerte forzamiento sinóptico, por lo que en su lugar deben emplearse enfoques novedosos como los algoritmos jerárquicos. En general, el modelo propuesto tiene la capacidad de producir pronósticos de energía eólica de alta calidad para la mayoría de las condiciones climáticas que prevalecen en esta región, y de- muestra un buen desempeño respecto a modelos similares para pronósticos de energía eólica a mediano plazo.

This study delves into the application of machine learning models for the early detection of Autism Spectrum Disorder (ASD). Early diagnosis and intervention are critical for improving the lives of individuals with ASD and their families. This research compares various machine learning models, including Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbors, and more, assessing their performance based on key metrics such as F1-Score, accuracy, precision, and recall. The study reveals the Multi-layer Perceptron (MLP) as the top-performing model with an impressive F1-Score of 79.35%, demonstrating its potential for accurate ASD detection. The feature importance analysis highlights the significant roles of gender, genetic predisposition, age at diagnosis, and ethnicity-related features in predicting ASD. This study underscores the promise of machine learning in ASD detection and emphasizes the importance of early intervention and personalized approaches to diagnosis.

Abstract University dropout has a detrimental impact on numerous students; this phenomenon may be associated with personal issues, economic constraints, and other factors. Given this situation, the importance of developing a predictive model for such cases arises. To achieve this, Machine Learning techniques were proposed and employed, including Logistic Regression, Naive Bayes, Multilayer Perceptron Neural Network, Decision Tree, Support Vector Machine, and Random Forest. A dataset was selected and underwent data cleaning, addressing missing values and outliers. Subsequently, records with the 'Enrolled' outcome variable were removed, focusing solely on 'Dropout' and 'Graduate' categories. Each model was trained and tested using cross-validation with folds. Ultimately, they were compared based on accuracy, precision, and recall metrics, leading to the conclusion that Logistic Regression is the technique that yields the best results for predicting university dropout in the considered dataset.

Resumen La deserción universitaria afecta negativamente a muchos estudiantes, este suceso puede estar relacionado con problemas personales, cuestiones económicas, entre otros. Ante tal situación surge la importancia de desarrollar una forma de predecir estos casos, para esto se propuso el uso de técnicas de Machine Learning, las utilizadas fueron Regresión Logística, Naive Bayes, Red Neuronal Perceptrón Multicapa, Árbol de Decisión, Support Vector Machine y Random Forest; se seleccionó un Dataset, que pasó por una limpieza de datos, se corrigieron los datos faltantes y los valores atípicos; luego se eliminaron los registros cuya variable de salida era Matriculado, centrándose en los tipos Abandono y Graduado. Cada modelo fue entrenado y probado mediante validación cruzada con pliegues, finalmente, se compararon en función de métricas de precisión, exactitud y exhaustividad, donde se concluyó que la Regresión Logística es la técnica que mejores resultados proporciona para predecir la deserción universitaria en el dataset considerado.

Abstract Delimitation of areas subject to flooding is crucial to understand water dynamics and fluvial changes. This study analyzed the potential of C-band Synthetic Aperture Radar (SAR) images acquired by the Sentinel-1 satellite in 2017, 2018, and 2019 to delineate flooded areas in the Central Amazon. The images were processed by the Artificial Neural Network Multi-Layer Perceptron (ANN-MLP) and two K-Nearest Neighbor (KNN-7 and KNN-11) machine learning (ML) classifiers. Pre-processing of Single Look Complex (SLC) SAR images involved the following methodological steps: orbit-file application; radiometric calibration (σ0); Range-Doppler terrain correction; speckle noise filtering; and conversion of linear data to backscattering coefficients (units in dB). We applied the Lee filter, with a window size of 3x3, for speckle filtering. A set of 6000 randomly distributed samples for training (70%), validation (20%), and test (10%) was obtained based on visual interpretation of Sentinel-2 optical satellite image acquired in the same years of SAR images. We found the largest flooded areas in 2019 in the study area (municipality of Parintins and Urucará, Amazonas River, Brazil): 6244km2 by the ANN-MLP classifier; 6268km2 by KNN-7; and 6290km2 by KNN-11, while the smallest flooded areas were found in 2018: 5364km2 by ANN-MLP; 5412km2 by KNN-7; and 5535km2 by KNN-11. The three classifiers presented Kappa coefficients between 0.77 and 0.91. ANN-MLP showed the best accuracy. The presence of shadow effects in the SAR images increased the commission errors.

Resumo A delimitação de áreas sujeitas a inundações é crucial para entender a dinâmica hídrica e as mudanças fluviais. Este estudo analisou o potencial de imagens de radar de abertura sintética (SAR) adquiridas na banda-C pelo satélite Sentinel-1 em 2017, 2018 e 2019 para delinear áreas inundadas na Amazónia Central. As imagens foram processadas pela Rede Neural Artificial Multi-Layer Perceptron (RNA-MLP) e dois classificadores de aprendizagem de máquina (ML) K-Nearest Neighbor (KNN-7 e KNN-11). O pré-processamento de imagens SAR Single Look Complex (SLC) envolveu as seguintes etapas metodológicas: aplicação do orbit-file; calibração radiométrica (σ0); correção de terreno Range-Doppler; filtragem de ruído speckle; e conversão de dados lineares para coeficientes de retroespalhamento (unidades em dB). O filtro de Lee com tamanho de janela de 3×3 foi aplicado para filtragem do ruído speckle. Um conjunto de 6000 amostras distribuídas aleatoriamente para treino (70%), validação (20%) e teste (10%) foi obtido com base na interpretação visual da imagem do satélite óptico Sentinel-2 adquiridas no mesmo ano das imagens de radar. As maiores áreas alagadas foram encontradas em 2019 na área de estudo (municípios de Parintins e Urucará, Rio Amazonas, Brasil): 6244km2 pelo classificador RNA-MLP; 6268km2 pelo KNN-7; e 6290km2 pelo KNN-11, enquanto as menores áreas alagadas foram encontradas em 2018: 5364km2 pelo classificador RNA -MLP; 5412km2 pelo KNN-7; e 5535km2 pelo KNN-11. Os três classificadores apresentaram coeficientes Kappa entre 0,77 e 0,91. A RNA-MLP apresentou a melhor precisão. A presença de efeitos de sombra nas imagens SAR aumentou os erros de comissão.

Résumé La délimitation de zones sujettes aux inondations est cruciale pour comprendre la dynamique hydrique et les changements fluviaux. Cette étude a analysé le potentiel des images radar à ouverture de synthétique (SAR) acquises en bande-C par le satellite Sentinel-1 en 2017, 2018 et 2019 pour délimiter les zones inondées en Amazonie Centrale. Les images ont été traitées par le Réseau de Neurones Artificiels Multicouches Perceptron (ARN-MLP) et deux classificateurs d'apprentissage automatique (ML) K-Nearest Neighbor (KNN-7 et KNN-11). Le prétraitement des images SAR images complexes à visée simple (SLC) a impliqué les étapes méthodologiques suivantes : application du fichier d'orbite; calibration radiométrique (σ0); correction de terrain Range-Doppler; filtrage du bruit de chatoiement; et conversion des données linéaires en coefficients de rétrodiffusion (unités en dB). Un filtre de Lee avec une taille de fenêtre de 3×3 a été appliqué pour filtrer le bruit de chatoiement. Un ensemble de 6000 échantillons assignés au hasard pour la formation (70%), la validation (20%) et les tests (10%) a été obtenu à partir de l'interprétation visuelle de l'image satellite optique Sentinel-2 acquise pour la même année que les images radar. Les plus grandes zones inondées ont été trouvées en 2019 dans la zone d'étude (municipalité de Parintins et Urucará, la Riviére Amazonas, Brésil) : 6244km2 selon le classificateur RNA -MLP; 6268km2 pour KNN-7; et 6290km2 pour KNN-11, alors que les plus petites zones inondées ont été trouvées en 2018: 5364km2 pour le classificateur RNA -MLP; 5412km2 pour KNN-7; et 5535km2 pour KNN-11. Les trois classifieurs présentaient des coefficients Kappa compris entre 0,77 et 0,91. L'ARN-MLP a montré une meilleure précision. La présence d'effets d'ombre dans les images SAR a augmenté les erreurs de commission.

Resumen La delimitación de áreas sujetas a inundaciones es crucial para comprender la dinámica del agua y los cambios fluviales. Este estudio analizó el potencial de las imágenes de radar de apertura sintética (SAR) adquiridas en la banda-C por el satélite Sentinel-1 en 2017, 2018 y 2019 para delinear áreas inundadas en la Amazonía Central. Las imágenes fueron procesadas por la Red Neuronal Artificial de Perceptrón Multicapa (RNA-MLP) y dos clasificadores de aprendizaje automático (ML) K-Nearest Neighbor (KNN-7 y KNN-11). El preprocesamiento de imágenes SAR complejas de una sola mirada (SLC) involucró los siguientes pasos metodológicos: aplicación del archivo de órbita; calibración radiométrica (σ0); corrección del terreno Range-Doppler; filtrado de ruido moteado; y conversión de datos lineales en coeficientes de retrodispersión (unidades en dB). Se aplicó El filtro Lee con un tamaño de ventana de 3×3 para filtrar el ruido moteado. Se obtuvo un conjunto de 6000 muestras asignadas aleatoriamente para entrenamiento (70%), validación (20%) y prueba (10%) en base a la interpretación visual de la imagen del satélite óptico Sentinel-2 adquirida el mismo año que las imágenes de radar. Los humedales más grandes se encontraron en 2019 en el área de estudio (municipio de Parintins y Urucará, Rió Amazonas, Brasil): 6244km2 por el clasificador RNA -MLP; 6268km2 por KNN-7; y, 6290km2 por KNN-11, mientras que, los humedales más pequeños se encontraron en 2018: 5364km2 por el clasificador RNA -MLP; 5412km2 por KNN-7; y 5535km2 por KNN-11. Los tres clasificadores presentaron coeficientes Kappa entre 0,77 y 0,91. RNA-MLP mostró la mejor precisión. La presencia de efectos de sombra en las imágenes SAR aumentó los errores de comisión.

Resumen Objetivo: Aplicar un modelo de análisis de redes neuronales artificiales (RNA) para identificar las variables que predicen el liderazgo asignado y el éxito académico en egresados de seis generaciones de la carrera de Medicina. Método: Estudio analítico, retrospectivo y comparativo. Participaron 1434 egresados. Se envió un cuestionario por correo electrónico que incluyó el consentimiento de participación voluntaria. Se realizó análisis estadístico multivariado mediante RNA del tipo perceptrón multicapa, árboles de decisión y análisis de impulsores. Resultados: Las RNA identificaron siete variables independientes que predijeron el éxito profesional y ocho para el liderazgo en los médicos egresados. Los árboles de decisión identificaron diferencias significativas en las variables desempeño profesional (p = 0.000), edad (p = 0.005) y actividades de educación continua (p = 0.034) relacionadas con el éxito profesional, y para el liderazgo las variables sexo (p = 0.000), promedio en el bachillerato (p = 0.042), realizar práctica clínica en el servicio social (p = 0.002) y actividades de educación continua (p = 0.011). Conclusiones: Las RNA identificaron las principales variables independientes predictoras del éxito profesional y el liderazgo de los egresados. El estudio abre dos líneas de investigación poco estudiadas con las técnicas de RNA en el área de la medicina.

Abstract Objective: To apply an artificial neural networks analysis (ANN) model to identify variables that predict assigned leadership and academic success in graduates of six generations of medical school. Method: Analytical, retrospective, comparative study. A total of 1434 graduates participated. A questionnaire was sent to them by e-mail including a voluntary participation consent. A multivariate statistical analysis using multi-layer perceptron ANN, decision trees and driver analysis was performed. Results: The ANN identified seven independent variables that predicted professional success and eight for leadership in medical graduates. The decision trees identified significant differences in the variables professional performance (p = 0.000), age (p = 0.005) and continuing education activities (p = 0.034) related to professional success, and for leadership the variables gender (p = 0.000), high school grades (p = 0.042), performing clinical practice during the social service year (p = 0.002) and continuing education activities (p = 0.011). Conclusions: The ANN identified the main independent predictor variables of professional success and leadership of the graduates. This study opens up two new lines of research little studied with the techniques of in the area of medicine.

ABSTRACT The ranking of seed lots is a fundamental process for all companies in the seed industry. This work aims to demonstrate data mining methods for ranking sorghum seed lots during the seed processing through analysis of quality control data. Germination and cold tests were performed to verify the physiological quality of the lots. Seed samples from each lot were evaluated in two moments: post-cleaning and finished product (ready for marketing). The results after pre-processing totaled 188 rows of data with six attributes, encompassing 150 lots accepted for marketing, 6 rejected, and 32 intermediate lots. The classifiers used were J48, Random Forest, Classification Via Regression, Naive Bayes, Multilayer Perceptron, and IBk. The Resample filter was used for adjustment of the data. The k-fold technique was used for training, with ten folds. The metrics of Accuracy, Precision, Recall, F-measure, and ROC Area were used to verify the accuracy of the algorithms. The results obtained were used to determine the best machine-learning algorithm. IBk and J48 presented the highest accuracy of data; the IBk technique presented the best results. The Resample filter was essential for solving the data imbalance problem. Sorghum seed lots can be classified with great accuracy and precision through artificial intelligence and machine learning technique.

RESUMO A classificação de lotes de sementes é um processo fundamental para todas as empresas do setor sementeiro. O objetivo do trabalho é demonstrar os métodos de mineração de dados de ranqueamento de lotes de sementes de sorgo durante o processo de beneficiamento, através de análises de dados do controle de qualidade. Os testes realizados foram germinação e teste de frio, com o objetivo de verificar a qualidade fisiológica dos lotes. As amostras de sementes de cada lote foram avaliadas em dois momentos: póslimpeza e produto acabado (pronto para comercialização). Os dados gerados, após o pré-processamento, totalizaram 188 linhas com seis atributos, contabilizando 150 lotes aceitos para comercialização, seis rejeitados e 32 denominados intermediários. Os classificadores utilizados foram J48, Random Forest, Classification Via Regression, Naive Bayes, Multilayer Perceptron e IBk. Utilizou-se o filtro Resample para ajustamento dos dados. A técnica empregada para treinamento foi a k-fold, com 10 folds. Para verificar a precisão dos algoritmos foram utilizadas as métricas de Acurácia, Precisão, Recall, F-measure e Área ROC. Com os resultados obtidos determinou-se o melhor algoritmo de aprendizagem de máquina. Verificou-se que o IBk e o J48 obtiveram maior acurácia nos dados, sendo que a técnica de IBk obteve o melhor resultado. O filtro Resample foi importante para resolver o problema do desequilíbrio dos dados. Concluímos ser possível classificar lotes de sementes de sorgo com grande acurácia e precisão através de inteligência artificial e sua técnica de aprendizado de máquina.

The objective of the present work was to apply self-organizing feature maps (SOFM) and multilayer perceptron (MLP) to evaluate the protective capacity of tert-butylhydroquinone (TBHQ), butylhydroxyanisole (BHA), and butylhydroxytoluene (BHT) antioxidants against the biodiesel oxidation reaction. For this, the antioxidant concentration and the biodiesel compliance parameters were used as continuous input variables and the type of antioxidant as a categorical. The SOFM proved to be an adequate tool for the classification of biodiesel samples containing antioxidants. The performance of the 5 most active networks of the model ranged from 89 to 99% for training, testing, and validation with an error below 1.42 × 10-4. Statistical tests applied to validate the model showed no significant difference between predicted and experimental values. The global sensitivity analysis showed that the relative protection factor (RPF) contributed with 34.89%, the antioxidant with 31.49%, the induction period with 10.69%, the water content with 6.00%, and the others all together with 16.93% in the construction of the regression models by MLP.

Abstract The present study applies computational intelligence techniques in the development of a hybrid model composed of Artificial Neural Networks (ANNs) and Genetic Algorithms (GAs) (MLP-GA) to estimate and fill in the gaps in the monthly variables of evaporation, maximum temperature and relative humidity to six regions in the state of Rio de Janeiro (RJ), Brazil. The results were evaluated using statistical techniques and compared with results obtained by the Multiple Linear Regression (RLM), Multilayer Perceptron (MLP) and Radial Basis Function (RBF) models and also compared with the data recorded by the weather stations. The correlation coefficient (r) between the evaporation estimates generated by MLP-GA with the recorded data showed a high relationship, remaining between 0.82 to 0.97. The average percentage error (MPE) ranged from 6.01% to 9.67%, indicating a accuracy between 90% to 94%. For the maximum temperature generated by MLP-GA the correlation with the recorded data remained between 0.97 to 0.99. It also presented the MPE between 0.95% to 1.57%, maintaining the accuracy of the estimated data between 98% to 99%. The correlation coefficient (r) between the relative humidity estimates generated with the MLP-GA remained between 0.89 a 0.97, the MPE between 1.15% to 1.89%, which guaranteed a rate higher than 98% of correctness in its estimates. Such results demonstrated gains in relation to the other applied models and allowed the accomplishment of the filling of most of the missing values.

Resumo O presente estudo aplica técnicas de inteligência computacional no desenvolvimento de um modelo híbrido composto por Redes Neurais Artificiais (RNAs) e Algoritmos Genéticos (AGs) (MLP-GA) para estimar e preencher lacunas nas variáveis mensais de evaporação, temperatura máxima e umidade relativa em seis regiões do estado do Rio de Janeiro (RJ), Brasil. Os resultados foram avaliados por meio de técnicas estatísticas e comparados com os resultados obtidos pelos modelos de Regressão Linear Múltipla (RLM), Perceptron de Multicamadas (MLP) e Redes de Função de Base Radial (RBF), além de serem comparados com os dados registrados pelas estações meteorológicas. O coeficiente de correlação (r) entre as estimativas de evaporação geradas pelo MLP-GA com os dados registrados mostrou uma relação elevada, permanecendo entre 0,82 e 0,97. O erro percentual médio (MPE) variou de 6,01% a 9,67%, indicando uma precisão entre 90% e 94%. Para a temperatura máxima gerada pelo MLP-GA, a correlação com os dados registrados permaneceu entre 0,97 e 0,99. Apresentou também o MPE entre 0,95% e 1,57%, mantendo a precisão dos dados estimados entre 98% e 99%. O coeficiente de correlação (r) entre as estimativas de umidade relativa geradas pelo MLP-GA permaneceu entre 0,89 e 0,97, com MPE entre 1,15% e 1,89%, garantindo uma taxa superior a 98% de acerto em suas estimativas. Tais resultados demonstraram ganhos em relação aos outros modelos aplicados e permitiram o preenchimento da maioria dos valores ausentes.

Abstract Artificial neural networks (ANNs) may experience problems due to insufficient or uninformative predictors, and these problems are common for complex predictions such as those for rainfall. However, some studies point to the use of climate variables and anomalies as predictors to make the forecast more accurate. This research aimed to predict the monthly rainfall, one month in advance, in four municipalities of the metropolitan region of Belo Horizonte using an ANN trained with different climate variables; additionally, it aimed to indicate the suitability of such variables as inputs to these models. The models were developed using the MATLAB® software Version R2011a using the NNTOOL toolbox. The ANNs were trained by the multilayer perceptron architecture and the feedforward and backpropagation algorithm using two combinations of input data, with two and six variables, and one combination of input data with the three most correlated variables to observed rainfall from 1970 to 1999 to predict the rainfall from 2000 to 2009. The climate variable most correlated with the rainfall of the following month was the average compensated temperature. Even when using the variables most correlated with precipitation as predictors (0.66 ≤ nt index ≤ 1.26), there was no notable improvement in the predictive capacity of the models when compared to those that did not use climate variables as predictors (0.55 ≤ nt index ≤ 0.80).

Resumo As redes neurais artificiais (RNAs) podem apresentar problemas devido a preditores insuficientes ou não informativos, o que é comum para previsões complexas, como as de precipitação pluvial. No entanto, alguns estudos apontam para o uso de variáveis e anomalias climáticas como preditores para tornar a previsão mais precisa. Esta pesquisa teve como objetivo prever a precipitação mensal, com um mês de antecedência, em quatro municípios da região metropolitana de Belo Horizonte utilizando uma RNA treinada com diferentes variáveis climáticas; além disso, buscou indicar a adequação de tais variáveis como entrada para esses modelos. Os modelos foram desenvolvidos por meio do software MATLAB® versão R2011a utilizando a toolbox NNTOOL. As RNAs foram treinadas pela arquitetura multilayer perceptron e pelo algoritmo feedforward e backpropagation usando duas combinações de dados de entrada, com duas e seis variáveis, e uma combinação de dados de entrada com as três variáveis mais correlacionadas com precipitação observada de 1970 a 1999 para prever a precipitação de 2000 a 2009. A variável climática mais correlacionada com a precipitação do mês seguinte foi a temperatura média compensada. Mesmo utilizando as variáveis mais correlacionadas com a precipitação como preditores (0,66 ≤ índice nt ≤ 1,26), não houve melhora significativa na capacidade preditiva dos modelos quando comparado aos que não utilizaram variáveis climáticas como preditores (0,55 ≤ índice nt ≤ 0,80).

ABSTRACT Objective: The variation in mortality rates of intensive care unit oncological patients may imply that clinical characteristics and prognoses are very different between specific subsets of patients with cancer. The specific characteristics of patients with cancer have not been included as risk factors in the established severity-of-illness scoring systems and comorbidity scores, showing limitations in predicting mortality risk. This study aimed to devise a predictive tool for in-hospital mortality for adult patients with a respiratory neoplasm admitted to the intensive care unit, using an artificial neural network. Methods: A total of 1,221 stays in the intensive care unit from the Beth Israel Deaconess Medical Center were studied. The primary endpoint was the all-cause in-hospital mortality prediction. An artificial neural network was developed and compared with six severity-of-illness scores and one comorbidity score. Model building was based on important predictors of lung cancer mortality, such as several laboratory parameters, demographic parameters, organ-supporting treatments, and other clinical information. Discrimination and calibration were assessed. Results: The AUROC for the multilayer perceptron was 0.885, while it was <0.74 for the conventional systems. The AUPRC for the multilayer perceptron was 0.731, whereas it was ≤0.482 for the conventional systems. The superiority of multilayer perceptron was statistically significant for all pairwise AUROC and AUPRC comparisons. The Brier Score was better for the multilayer perceptron (0.109) than for OASIS (0.148), SAPS III (0.163), and SAPS II (0.154). Conclusion: Discrimination was excellent for multilayer perceptron, which may be a valuable tool for assessing critically ill patients with lung cancer.

Resumen La frontera que comparten México y Guatemala es una región compleja de analizar por la diversidad de dinámicas entre ambos países. Una amenaza de la región es el cambio de uso de suelo, donde existen una serie de impulsores de cambio y condicionantes ambientales que hacen prioritario su análisis. La presente investigación revisa los cambios de uso de suelo que ocurrieron entre 1990 y 2010, mismos que se usan como base para construir un escenario tendencial al año 2030 a través de utilizar la red neuronal perceptrón multicapa. El escenario resultante muestra el crecimiento de la agricultura y de las zonas urbanas, así como la pérdida de bosques, otras vegetaciones y cuerpos de agua. Esta información evidencia la necesidad de analizar las implicaciones socioambientales que ello podría traer consigo.

Abstract The shared border between Mexico and Guatemala is a complex region to analyze due to the diverse dynamics between both countries. One threat to the area is the land-use change, where many drivers of change and environmental conditions make it a priority to analyze. This research studies the land-use changes that occurred between 1990 and 2010; these are used as a basis for constructing a trend scenario to the year 2030 through the use of the multilayer perceptron neural network. The resulting scenario shows the growth of agriculture and urban areas, as well as the loss of forests, other vegetation, and water bodies. This information highlights the need to analyze the socio-environmental implications that this could bring about.

RESUMEN La estimación precisa de la evapotranspiración (ET) es esencial para gestionar el riego en cultivos, pero no es una tarea fácil. Las metodologías empíricas de ET requieren mediciones precisas de la radiación neta (Rn) para obtener resultados confiables. Sin embargo, estas mediciones no son rutinarias en las estaciones meteorológicas. Este trabajo exploró el uso de aprendizaje automático para estimar la ET diaria con dos sustitutos de Rn: la radiación solar extraterrestre (Ra) y la Rn modelada (RnM). Se utilizó Support Vector Machine (SVM), Kernel Ridge (KR), Decision Tree (DT), Adaptive Boosting (AB) y Multilayer Perceptron (MLP) para modelar observaciones de FLUXNET. Adaptive Boosting brindó los mejores resultados con observaciones de Rn (RnO), con un valor para la raíz del error cuadrático medio de aproximadamente el 16 % de Rn medio observado. La Rn resultante (AB RnM) se utilizó para modelar la ET, usando RnO, AB RnM y Ra, junto a variables meteorológicas y el índice NDVI. Los métodos evaluados estimaron adecuadamente la ET, arrojando errores similares a los obtenidos con RnO, cuando se contrastan con las observaciones de ET. Estos resultados demuestran que AB y KR son aplicables con datos rutinarios meteorológicos y de satélite para estimar la ET.

SUMMARY Accurate evapotranspiration (ET) estimation is essential for water management in crops, but it is not an easy task. Empirical ET methodologies require precise net radiation (Rn) measurements to obtain accurate results. Nevertheless, Rn measurements are not easy to obtain from meteorological stations. Thus, this study explored the use of machine learning algorithms with two Rn substitutes, to estimate daily ET: the extraterrestrial solar radiation (Ra) and a modelled Rn (RnM). Support Vector Machine (SVM), Kernel Ridge (KR), Decision Tree (DT), Adaptive Boosting (AB), and Multilayer Perceptron (MLP) were applied to model FLUXNET Rn and ET observations. Adaptive Boosting produced the best field Rn measurements (RnO), yielding a Root Mean Square Error of about 16 % of the mean observed Rn. The resulting Rn (AB RnM) was used to model daily crops ET employing the above-mentioned machine learning methods with RnO, AB RnM, and Ra, in conjunction with meteorological variables and the NDVI index. The evaluated methods were suitable to estimate ET, yielding similar errors to those obtained with RnO, when contrasted with ET observations. These results demonstrate that AB and KR are applicable with rutinary meteorological and satellite data to estimate ET.

RESUMO Este trabalho teve como objetivo desenvolver um novo método híbrido de calibração de rugosidades absolutas em redes de distribuição de água, com base na combinação do método iterativo do gradiente hidráulico alternativo (Migha) com uma rede neural artificial (RNA) do tipo Perceptron Multicamadas. Foram testadas diferentes configurações para o modelo, em vários cenários que diferiam quanto ao número de pressões observadas utilizadas. Os resultados encontrados sugerem que a combinação dos métodos foi favorável à calibração das rugosidades, com erros absolutos médios inferiores a 0,016 mm no modelo que calculava apenas um valor por material para o parâmetro. A boa performance também ficou evidente quando analisadas as pressões e vazões retornadas pela simulação da rede calibrada, as quais apresentaram erros relativos médios de 0,12 e 0,36%, respectivamente, no cenário com 14% dos nós com pressões conhecidas, erros consideravelmente inferiores àqueles obtidos pela configuração inicial da rede e pelo uso exclusivo do Migha.

ABSTRACT This work aimed to develop a new hybrid method for the calibration of absolute roughness in water distribution networks, starting from the combination of the Alternative Hydraulic Gradient Iterative Method (Migha) with a Multilayer Perceptron Artificial Neural Network (ANN). Different configurations were tested for the model, in different scenarios that differed in the number of observed pressures used. The results found suggest that the combination of methods was favorable to roughness calibration, with mean absolute errors lower than 0.016 mm in the model that calculated only one value per material for the parameter. The good performance was also evident when analyzing the pressures and flows returned by the calibrated network simulation, which showed average relative errors of 0.12 and 0.36%, respectively, in the scenario with 14% of nodes with known pressures, errors considerably lower than those obtained by the initial configuration of the network and the exclusive use of Migha.

Biodiesel is capable of replacing diesel because it has similar physicochemical properties, but this biofuel is susceptible to oxidation, which makes the application of antioxidant substances necessary. For this study, alcoholic extracts of senna leaves, hibiscus flowers, and blackberry were used. Biodiesel samples were submitted to physicochemical analysis to evaluate interference in the volume of these alcoholic extracts with antioxidant properties. The data obtained were processed using the neural network of the multilayer perceptron type (MLP). For the network’s training, 200 epochs were used. The samples were randomly divided into three groups, with 70% used for training, 15% for testing, and 15% for validation. The type of extract was considered as a categorical variable, the extract volume as a target variable, and the other ones as input variables. Among the 200 networks trained, with 5 to 20 hidden layers, the 5 with the best performance were highlighted. The Tukey test applied to the means showed no significant difference at the 5% level, between the value of the added volume and the means value predicted by the networks. The sensitive analysis showed that the most important input variable for the construction of the model was the type of extract.

Abstract In this research, two machine learning classifiers were implemented, a multilayer perceptron (MLP) and a gradient boosting model (GB), to predict the degree of academic achievement in Spanish and mathematics of basic education students in two stages, sixth of primary (2008) and third of secondary (2011), based on contextual variables obtained from the Enlace test of the state of Tlaxcala, Mexico. Thirteen input variables were considered. The relative importance of these was determined by the random forest (RF) classifier. MLP and GB classifiers were trained and tested with a dataset of 11 036 records of students who remained in the school system from 2008 to 2011. The models were trained and tested in prediction for 2008 and 2011. In Spanish MLP outperformed GB with a global classification accuracy (PG) of 70.1 % in 2008 and 61.1 % in 2011. GB obtained better performance in mathematics with a PG of 68.8 % in 2008 and 63.5 % in 2011. It was observed that the score in Spanish has a strong association with the degree of academic achievement in mathematics. Scores in Spanish and mathematics have greater relative importance with respect to contextual factors considered as sex, scholarship, school shift, and so on. In the population of students analyzed, it is observed that, in Spanish and mathematics, the proportion of women is higher than the proportion of men in achievement levels 1 (elementary) and 2 (good or excellent); in contrast, in both subjects this proportion is reversed at achievement level 0 (insufficient).

Resumen En esta investigación se implementaron dos clasificadores de aprendizaje automático, una red neuronal multicapa (perceptrón multicapa [MLP]) y un modelo de potenciación del gradiente (GB), para predecir el grado de logro académico en las asignaturas de español y matemáticas de alumnos de sexto de primaria (2008) y tercero de secundaria (2011) con base en variables contextuales obtenidas de los Exámenes Nacionales del Logro Académico en Centros Escolares (Enlace) del estado de Tlaxcala, México. Se consideraron 13 variables de entrada y la importancia relativa de éstas, se determinó por medio del algoritmo bosque aleatorio (RF). Los clasificadores MLP y GB se entrenaron y probaron con un conjunto de datos de 11 036 registros de estudiantes que permanecieron en el sistema escolar de 2008 a 2011. Los modelos se entrenaron y probaron en predicción para 2008 y 2011. En español MLP fue superior a GB con una precisión global de clasificación (PG) de 70.1 % en 2008 y 61.1 % en 2011. GB obtuvo mejores resultados en matemáticas con una PG de 68.8 % en 2008 y 63.5 % en 2011. Se observó que el puntaje en español tiene una fuerte asociación con el grado de logro académico en matemáticas. Los puntajes en español y matemáticas tuvieron mayor importancia relativa con respecto a los factores contextuales analizados como: sexo, beca, turno de la escuela. En la población de alumnos analizada se observó que en español y matemáticas la proporción de mujeres es mayor a la proporción de hombres en los grados de logro académico elemental, y bueno o excelente; en contraste, en ambas asignaturas esta proporción se invierte con el grado de logro insuficiente.

Resumo Nesta pesquisa, dois classificadores de aprendizado de máquina, uma rede neural multicamada (multilayer perceptron [MLP]) e um modelo de potenciação de gradiente (GB), foram implementados para prever o grau de desempenho acadêmico nas disciplinas de espanhol e matemática de alunos do ensino médio. sexta série (2008) e terceira série (2011) com base em variáveis contextuais obtidas nos Exames Nacionais de Desempenho Acadêmico nas Escolas (Enlace) do estado de Tlaxcala, México. 13 variáveis de entrada foram consideradas e sua importância relativa foi determinada usando o algoritmo Random Forest (RF). Os classificadores MLP e GB foram treinados e testados com um conjunto de dados de 11.036 prontuários de alunos que permaneceram na rede escolar de 2008 a 2011. Os modelos foram treinados e testados em previsão para 2008 e 2011. Em espanhol, o MLP foi superior ao GB com uma precisão geral de notas (GP) de 70,1% em 2008 e 61,1% em 2011. GB teve um desempenho melhor em matemática com um GP de 68,8% em 2008 e 63,5% em 2011. A pontuação em espanhol mostrou ter uma forte associação com o grau de desempenho acadêmico em matemática. Os escores em espanhol e matemática tiveram maior importância relativa em relação aos fatores contextuais analisados como: gênero, escolaridade, turno escolar. Na população de alunos analisada, observou-se que em espanhol e matemática a proporção de mulheres é maior do que a proporção de homens no ensino fundamental e nas notas de desempenho acadêmico bom ou excelente; e essa proporção se inverte com o grau de realização insuficiente.