The increase in the average temperature of the Earth due to excess CO2 and other greenhouse gases has caused glaciers to melt and sea levels to rise. The burning of fossil fuels is one of the main causes of this problem, so the use of biofuels is important for mitigating this environmental impact. In Brazil, biodiesel is mixed with mineral; however, irregularities have been detected, reducing the expected positive effects. Therefore, this study proposes the use of multivariate curve resolution with alternating least squares (MCR ALS) to quantify the content of methyl palm oil biodiesel in diesel blends and determine the concentration of the adulterant automotive lubricating oil in the blend. The model for quantifying biodiesel in diesel obtained R2 = 0.9997, root mean square error of calibration (RMSEC) = 0.56% (v v-1) and prediction (RMSEP) = 0.65% (v v-1), while the determination of adulteration in the biodiesel-diesel blend showed RMSEC = 0.67% (v v-1), RMSEP = 0.81% (v v-1) and R2 = 0.9992. These results demonstrate the ability of the models developed to predict both the content and quantity of possible adulterants in blends of palm oil biodiesel and mineral diesel, contributing to the quality control of this fuel. CO rise problem impact Brazil however detected effects Therefore MCR ALS R 09997 0 9997 0.9997 (RMSEC 056 56 0.56 v v1 1 v-1 (RMSEP 065 65 0.65 v1, , biodieseldiesel 067 67 0.67 081 81 0.81 09992 9992 0.9992 fuel 0999 999 0.999 05 5 0.5 v- 06 6 0.6 08 8 0.8 099 99 0.99 0. 09 9 0.9

Livestock farms in Kenya face pressure to increase productivity amid rising global population. Cattle farming dominates, but small to medium-sized farms struggle with cattle insemination. Currently, visual observation is used for heat detection, with farmers maintaining farm journals. Modern methods utilizing sensors to improve estrus prediction are time-consuming, costly and need constant internet connection. This research proposes a novel approach-the use of an on-controller machine learning algorithm- for estrus prediction in cattle. Motion and temperature data was collected from two zero-grazed multiparous Holstein Friesian cows in Kiambu County, Kenya for 11 months. The data was cleaned and stored. Movement intensity profiles were derived by root-mean-squaring directional accelerometer values and averaging this over time. Validation was performed by observing cow behavior for indicators such as restlessness, mounting, and vulva swelling, with farmer predictions documented in their records. The collected data was then used to train a machine learning algorithm, with several models tested, and a neural network emerged as the best fit. The TensorFlow library facilitated the implementation of the algorithm on a microcontroller, allowing for the development of an animal tag featuring the ML algorithm. Results demonstrated 83.9% sensitivity, 89.0% specificity and 89.5% accuracy in detecting estrus, compared to farmer's visual observation, which had only 37% sensitivity. These findings underscore the potential to integrate machine learning into Precision Livestock Farming for estrus prediction, with prediction occurring directly on the animal tag offline. This integration holds promise for farmers, notably heightened insemination success rates, without necessitating significant financial investment.

RESUMEN Introducción: La validación de métodos de predicción de riesgo es importante para establecer medidas preventivas adecuadas. Objetivo: Evaluar la validez interna de una escala de mortalidad por afección cardiovascular en ancianos. Métodos: Se efectuó un estudio analítico, observacional, retrospectivo, de tipo caso-control, de 58 pacientes fallecidos (casos) y 134 vivos (controles), pertenecientes a 3 áreas de salud (policlínicos Ramón López Peña, Municipal y 28 de septiembre) del municipio de Santiago de Cuba, durante el 2021. Para realizar la validación de dicha escala se evaluó la capacidad discriminativa y la calibración; la validez de apariencia, criterio y constructo, así como su fiabilidad y consistencia interna. Resultados: La escala diseñada discriminó mejor que la EPICARDIAN en ambos sexos; 68,8 % de los fallecidos del grupo de validación fueron clasificados de alto riesgo. El valor del coeficiente alfa de Cronbach fue de 0,639 y el estandarizado de 0,564. Conclusiones: La escala evaluada calibra adecuadamente al aplicarla en otra población con similar característica a la que le dio origen y discrimina de manera pertinente a las personas clasificadas como alto riesgo de fallecer por una enfermedad cardiovascular aterosclerótica.

ABSTRACT Introduction: The validation of risk prediction methods is important to establish appropriate preventive measures. Objective: To evaluate the internal validity of a mortality scale due to cardiovascular affection in elderly. Methods: An analytic, observational, retrospective and case-control study was carried out, in 58 dead patients (cases) and 134 alive patients (control), belonging to 3 health areas (Ramón López Peña, Municipal and 28 de Septiembre polyclinics) in Santiago de Cuba municipality, during 2021. To validate this scale the discriminative capacity and calibration; appearance, criterion and construct validity were evaluated, as well as its reliability and internal strength. Results: The designed scale discriminated better than the EPICARDIAN scale in both sexes; 68.8 % of the deaths in the validation group were classified as high risk. The value of Cronbach´s alpha coefficient was 0.639 and the standardized one was 0.564. Conclusions: The evaluated scale calibrates appropriately when applied to another population with similar characteristics to the one it gave rise to and it pertinently discriminates against people classified as high risk of dying due to atherosclerotic cardiovascular disease.

Economic transformation and escalating market competitiveness have prompted manufacturers to adopt zero-defect manufacturing principles to lower production costs and maximise product quality. The key enabler of zero-defect manufacturing is the adoption of data-driven techniques that harness the wealth of information offered by digitalised manufacturing systems in order to predict errors. Multi-stage manufacturing systems, however, introduce additional complexity owing to the cascade effects associated with stage interactions. A generic modular framework is proposed for facilitating the tasks associated with preparing data emanating from multi-stage manufacturing systems, building predictive models, and interpreting these models' results. In particular, cascade quality prediction methods are employed to harness the benefit of invoking a stage-wise modelling approach. The working of the framework is demonstrated in a practical case study involving data from a multistage semiconductor production process.

Ekonomiese transformasie en toenemende markmededingendheid het vervaardigers aangespoor om zero-defek-vervaardigingsbeginsels toe te pas met die oog om produksiekoste te verlaag en produkkwaliteit te maksimeer. Die sleutelbemiddelaar van zero-defek-vervaardiging is die gebruik van data-gedrewe tegnieke om die magdom inligting wat deur gedigitaliseerde vervaardigingstelsels beskikbaar gemaak word, vir die voorspelling van foute te benut. Multi-stadium vervaardigingstelsels veroorsaak egter bykomende kompleksiteit as gevolg van kaskade-effekte wat verband hou met stadiuminteraksies. 'n Generiese modulêre raamwerk word voorgestel vir die fasilitering van take wat verband hou met die voorbereiding van data wat uit multi-stadium vervaardigingstelsels voortspruit, die bou van voorspellende modelle en die interpretasie van hierdie modelresultate. In die besonder word metodes vir die voorspelling van kaskadekwaliteit toegepas om die voordeel van 'n stadium-gewyse modelleringsbenadering uit te buit. Die werking van die raamwerk word in 'n praktiese gevallestudie gedemonstreer wat betrekking het op 'n multi-stadium halfgeleier produksie proses.

RESUMEN: El presente estudio tuvo como objetivo investigar la utilidad del fémur proximal en la estimación forense de la edad mediante la evaluación de cambios en las densidades óseas a través de radiografías. Utilizando el umbral de Otsu, la densidad ósea se cuantificó contando todos los valores de pixeles blancos dentro de regiones de interés seleccionadas, que incluyen la cabeza femoral (CF), el cuello femoral (CF), el triángulo de Ward (WT) y el trocánter mayor (TM) de 354 fémures izquierdos de ascendencia del norte de Tailandia. También se estimó el ancho de pixeles de la cavidad medular (CM). Además, el estudio evaluó el rendimiento de modelos de regresión lineal (RL) para la estimación de la edad a partir de imágenes radiográficas de fémur proximal. Se observaron correlaciones negativas entre la intensidad de los pixeles CF, CF, WT y TM con la edad de muerte, y las mujeres exhibieron correlaciones más fuertes que los hombres. Además, se encontró una correlación positiva entre la edad y el ancho del CM en muestras de mujeres, mientras que el ancho del CM del hombre no mostró ninguna relación con el aumento de la edad. Los resultados mostraron una ligera diferencia entre el modelo RL aplicado a ambos sexos, que integraba todas las variables, y la configuración alternativa que sólo utilizaba atributos relevantes. Ambos modelos mostraron un rendimiento similar, con un rango estrecho de valores del error cuadrático medio (RMSE), que oscilaba entre 12,67 y 12,71 años, y un rango de coeficiente de correlación de 0,51 a 0,52. Para las mujeres, el modelo RL con CF y WT como atributos seleccionados (RMSE = 11,85 años, coeficiente de correlación = 0,65) tuvo un desempeño satisfactorio, mientras que para los hombres, el modelo RL con todas las variables mostró un RMSE de 12,52 años y un coeficiente de correlación de 0,46. Este estudio mostró la posible aplicación de la intensidad de los pixeles en la predicción de la edad.

SUMMARY: The present study aimed to investigate the utility of the proximal femur in the forensic age estimation by assessing changes in bone densities through radiographs. Using Otsu’s threshold, bone density was quantified by counting all white pixel values within selected regions of interest, which include femoral head (FH), femoral neck (FN), Ward’s triangle (WT), and greater trochanter (GT) from 354 left femora of Northern Thai descent. The pixel width of medullary cavity (MC) was also estimated. Furthermore, the study evaluated the performance of linear regression (LR) models for age estimation from radiographic images of proximal femora. Negative correlations were observed between FH, FN, WT, and GT pixel intensity with the age-at-death of the samples, with females exhibiting stronger correlations than males. Moreover, a positive correlation was found between age and MC width in female samples, while male MC widths did not show any relationship with increasing age. The results showed a slight difference between the LR model applied to both sexes, which integrated all variables, and the alternative configuration that only utilized relevant attributes. Both models exhibited similar performance, with a narrow range of root mean square error (RMSE) values, ranging from 12.67 to 12.71 years, and a correlation coefficient range of 0.51 to 0.52. For females, the LR model with FN and WT as selected attributes (RMSE = 11.85 years, correlation coefficient = 0.65) performed decently, while for males, the LR model with all variables showed RMSE of 12.52 years and correlation coefficient of 0.46. This study showcased the potential application of pixel intensity in predicting age.

Abstract: The primary objectives of this research study are to provide a detailed description of machine learning (ML) technology when applied to assessing innovation and to design a model that allows predicting an institution’s degree of innovation. Machine learning technology lacks assumptions or preconceptions and is capable of processing a large amount of data and variables. After data processing, the ML model is built using variables associated with educational context, training is performed, and a web is built to predict the degree of innovation of an educational institution. This model provides an innovation accuracy prediction of 66% and allows assessing the influence of the variables analyzed when predicting the use of active methodologies at a given institution. In conclusion, this approach can open new data analysis techniques supported by ML that complement traditional statistically based approaches.

Resumen: Los principales objetivos de esta investigación son proporcionar una descripción detallada de la tecnología de aprendizaje automático (ML, por sus siglas en inglés) aplicada a la medición de la innovación y diseñar un modelo que permita predecir el grado de innovación en una determinada institución. La tecnología ML carece de suposiciones o preconcepciones y es capaz de procesar una gran cantidad de datos y variables. Tras procesar los datos, se construye el modelo ML empleando variables asociadas al contexto educativo, se realiza el entrenamiento y se construye una web para predecir el grado de innovación de una institución educativa. El modelo proporciona una precisión del 66% en la predicción de la innovación y permite discutir la influencia de las variables analizadas a la hora de predecir el uso de metodologías activas en una institución. En conclusión, este enfoque puede abrir nuevas técnicas de análisis de datos apoyadas en ML que complementen los enfoques tradicionales basados en la estadística.

Abstract Objective: Ovarian torsion (OT) represents a severe gynecological emergency in female pediatric patients, necessitating immediate surgical intervention to prevent ovarian ischemia and preserve fertility. Prompt diagnosis is, therefore, paramount. This retrospective study set out to assess the utility of combined clinical, ultrasound, and laboratory features in diagnosing OT. Methods: The authors included 326 female pediatric patients aged under 14 years who underwent surgical confirmation of OT over a five-year period. Logistic regression analysis was employed to pinpoint factors linked with OT, and the authors compared clinical presentation, laboratory results, and ultrasound characteristics between patients with OT (OT group) and without OT (N-OT group). The authors conducted receiver operating characteristic (ROC) curve analysis to gauge the predictive capacity of the combined features. Results: Among 326, OTwas confirmed in 24.23 % (79 cases) of the patients. The OT group had a higher incidence of prenatal ovarian masses than the N-OT (22 cases versus 7 cases) (p < 0.0001). Similarly, the authors observed significant differences in the presence of lower abdominal pain, suspected torsion on transabdominal ultrasound, and a high neutrophil-lymphocyte ratio (NLR > 3) between the OTand non-OT groups (p < 0.05). Furthermore, when these parameters were combined, the resulting area under the curve (AUC) was 0.868, demonstrating their potential utility in OT diagnosis. Conclusion: This study demonstrates a prediction model integrating clinical, laboratory, and ultrasound findings that can support the preoperative diagnosis of ovarian torsion, thereby enhancing diagnostic precision and improving patient management. Future prospective studies should concentrate on developing clinical predictive models for OTin pediatric patients. Objective fertility is therefore paramount Methods 32 1 fiveyear five year period presentation results NOT N group. . ROC (ROC Results 2423 24 23 24.2 79 (7 22 (2 p 0.0001. 00001 0.0001 0 0001 0.0001) Similarly pain neutrophillymphocyte neutrophil lymphocyte NLR 3 nonOT non 0.05. 005 0.05 05 0.05) Furthermore AUC (AUC 0868 868 0.868 Conclusion management 242 2 24. ( 0000 0.000 000 00 0.0 086 86 0.86 0.00 0. 08 8 0.8

Energy consumption from the residential sector forms a large portion of the electricity grid demand. The growing accessibility of residential load profile data presents an opportunity for improved residential load forecasting and the implementation of demand-side management (DSM) strategies. Machine learning is a tool well-suited for predicting stochastic processes, such as residential power usage due to human behavior. Long short-term memory (LSTM) recurrent neural networks are especially suited for predicting time-series data such as electrical load profiles. This paper investigates the impact of LSTM hyperparameters to the predictive performance of models, which include the tradeoffs associated with training data size, horizon ratios, model fidelity, prediction horizon and computational intensity. This paper provides a framework to evaluate the choice of LSTM hyperparameters for understanding trade-offs in a practical application of load profile predictions for the context of Grid-interactive Efficient Buildings (GEBs).

Abstract Land value patterns show very distinct spatial associations with accessibility to urban centralities and physical factors in a territory. However, predictions based on models of this structure can be highly uncertain, as the underlying data also may show clustering (thus allowing for better predictions in more densely sampled areas). An assessment of this uncertainty for land value extrapolations in the San José Metropolitan Region of Costa Rica is presented, via conditional Gaussian simulation, and the determinants of this uncertainty were explored, to find spatial strengths and weaknesses in the modeling efforts. The E-Type prediction from the conditional Gaussian simulation was found to marginally improve on ordinary kriging methods and it also provided explicit uncertainty patterns, which are the inverse of the land value prediction. The estimated uncertainty was found to decrease with characteristics that identify suitability for urban land use (and thus higher land values).

Resumen Los patrones de valor del suelo muestran asociaciones espaciales claras con accesibilidad a centralidades urbanas y a factores físicos de un territorio. Sin embargo, las predicciones basadas en esta estructura pueden ser altamente inciertas, dado que los datos mismos también exhiben aglomeración (y, por tanto, permiten mejores predicciones en las zonas más densamente muestreadas). Se presenta una evaluación de esta incertidumbre para extrapolaciones de valor del suelo en la Gran Área Metropolitana de Costa Rica mediante simulaciones gaussianas condicionales y una exploración de los determinantes de esta incertidumbre, como forma de reconocer fortalezas y debilidades de esta predicción. La predicción E-Type simulada resultó marginalmente mejor que extrapolaciones mediante kriging ordinario y produjo una cuantificación espacialmente explícita de la incertidumbre. El patrón de incertidumbre resultó ser un espejo de los valores del suelo. Se encontró que la incertidumbre se reduce con características asociadas a mayor aptitud del suelo para usos urbanos y, por tanto, de mayor precio.

Abstract Objective: Reliably prediction models for coronary artery abnormalities (CAA) in children aged > 5 years with Kawasaki disease (KD) are still lacking. This study aimed to develop a nomogram model for predicting CAA at 4 to 8 weeks of illness in children with KD older than 5 years. Methods: A total of 644 eligible children were randomly assigned to a training cohort (n = 450) and a validation cohort (n = 194). The least absolute shrinkage and selection operator (LASSO) analysis was used for optimal predictors selection, and multivariate logistic regression was used to develop a nomogram model based on the selected predictors. Area under the receiver operating characteristic curve (AUC), calibration curves, Hosmer-Lemeshow test, Brier score, and decision curve analysis (DCA) were used to assess model performance. Results: Neutrophil to lymphocyte ratio, intravenous immunoglobulin resistance, and maximum baseline z-score ≥ 2.5 were identified by LASSO as significant predictors. The model incorporating these variables showed good discrimination and calibration capacities in both training and validation cohorts. The AUC of the training cohort and validation cohort were 0.854 and 0.850, respectively. The DCA confirmed the clinical usefulness of the nomogram model. Conclusions: A novel nomogram model was established to accurately assess the risk of CAA at 4–8 weeks of onset among KD children older than 5 years, which may aid clinical decisionmaking. Objective (CAA (KD lacking Methods 64 n 450 194. 194 . 194) (LASSO AUC, , (AUC) curves HosmerLemeshow Hosmer Lemeshow test score (DCA performance Results ratio resistance zscore z 25 2 2. cohorts 0854 0 854 0.85 0850 850 0.850 respectively Conclusions 48 4– decisionmaking 6 45 19 (AUC 085 85 0.8 1 08 0.

Abstract Objective: Periventricular-intraventricular hemorrhage is the most common type of intracranial bleeding in newborns, especially in the first 3 days after birth. Severe periventricular-intraventricular hemorrhage is considered a progression from mild periventricular-intraventricular hemorrhage and is often closely associated with severe neurological sequelae. However, no specific indicators are available to predict the progression from mild to severe periventricular-intraventricular in early admission. This study aims to establish an early diagnostic prediction model for severe PIVH. Method: This study was a retrospective cohort study with data collected from the MIMIC-III (v1.4) database. Laboratory and clinical data collected within the first 24 h of NICU admission have been used as variables for both univariate and multivariate logistic regression analyses to construct a nomogram-based early prediction model for severe periventricular-intraventricular hemorrhage and subsequently validated. Results: A predictive model was established and represented by a nomogram, it comprised three variables: output, lowest platelet count and use of vasoactive drugs within 24 h of NICU admission. The model’s predictive performance showed by the calculated area under the curve was 0.792, indicating good discriminatory power. The calibration plot demonstrated good calibration between observed and predicted outcomes, and the Hosmer-Lemeshow test showed high consistency (p = 0.990). Internal validation showed the calculated area under a curve of 0.788. Conclusions: This severe PIVH predictive model, established by three easily obtainable indicators within the NICU, demonstrated good predictive ability. It offered a more user-friendly and convenient option for neonatologists. Objective Periventricularintraventricular Periventricular intraventricular newborns birth periventricularintraventricular periventricular sequelae However Method MIMICIII MIMIC III v1.4 v14 v v1 4 (v1.4 database 2 nomogrambased nomogram based validated Results output models s 0792 0 792 0.792 power outcomes HosmerLemeshow Hosmer Lemeshow p 0.990. 0990 0.990 . 990 0.990) 0788 788 0.788 Conclusions ability userfriendly user friendly neonatologists v1. (v1. 079 79 0.79 099 0.99 99 078 78 0.78 (v1 07 7 0.7 09 0.9 9 (v 0.

Abstract Objective: To build a model based on cardiometabolic indicators that allow the identification of overweight adolescents at higher risk of subclinical atherosclerotic disease (SAD). Methods: Cross-sectional study involving 161 adolescents with a body mass index ≥ + 1 z-Score, aged 10 to 19 years. Carotid intima-media complex thickness (IMT) was evaluated using ultrasound to assess subclinical atherosclerotic disease. Cardiometabolic indicators evaluated included nutritional status, central adiposity, blood pressure, lipidic profile, glycemic profile, as well as age and sex. Data was presented using measures of central tendency and dispersion, as well as absolute and relative frequency. The relationship between IMT measurement (outcome variable) and other variables (independent variables) was assessed using Pearson or Spearman correlation, followed by multiple regression modeling with Gamma distribution to analyze predictors of IMT. Statistical analysis was performed using SPSS and R software, considering a significance level of 5 %. Results: It was observed that 23.7 % had Carotid thickening, and the prevalence of abnormal fasting glucose was the lowest. Age and fasting glucose were identified as predictors of IMT increase, with IMT decreasing with age by approximately 1 % per year and increasing with glucose by around 0.24 % per mg/dL. Conclusion: The adolescent at higher risk is younger with higher fasting glycemia levels. Objective SAD. SAD . (SAD) Methods Crosssectional Cross sectional 16 zScore, zScore z Score, Score z-Score years intimamedia intima media (IMT status adiposity pressure profile sex dispersion frequency outcome variable independent correlation software Results 237 23 7 23. thickening lowest increase 024 0 24 0.2 mgdL mg dL mg/dL Conclusion levels (SAD 2 02 0.

Resumo O Brasil enfrenta altas taxas de inadimplência, devido em parte à pandemia, levando à busca de novas estratégias de cobranças de dívidas. O Machine Learning (ML), empregado com sucesso em inúmeras áreas, é aliado para elevar a eficácia dessas operações. Este artigo busca apresentar um panorama atual das pesquisas sobre aplicações de ML nas operações de cobrança de dívidas, por meio de uma Revisão Sistemática da Literatura. Foi utilizada a metodologia PICO, identificando inicialmente 41 documentos, dos quais 11 passaram por revisão sistemática. Os resultados mostraram quatro objetivos buscados pelos estudos: previsão de inadimplência, personalização das estratégias de cobrança, otimização das ações de recuperação de dívidas e previsão de recuperação de crédito. E os principais algoritmos utilizados foram Decision Tree, Logistic Regression, Random Forest, Naive Bayes, Artificial Neural Network e Deep Learning. Os resultados revelaram que ML é ainda pouco explorado nessa área, oferecendo potencial para avanços substanciais das pesquisas.

Abstract Brazil is facing high default rates, due in part to the pandemic, leading to the search for new debt collection strategies. Machine Learning (ML), successfully used in numerous areas, is an ally to increase the effectiveness of these operations. This article seeks to present a current overview of research on ML applications in debt collection operations, through a Systematic Literature Review. The PICO methodology was used, initially identifying 41 documents, of which 11 underwent systematic review. The results showed four objectives pursued by the studies: default prediction, personalization of collection strategies, optimization of debt recovery actions and credit recovery prediction. And the main algorithms used were Decision Tree, Logistic Regression, Random Forest, Naive Bayes, Artificial Neural Network and Deep Learning. The results revealed that ML is still little explored in this area, offering potential for substantial research advances.

ABSTRACT The aim of this study was to develop visible and near-infrared spectroscopy (Vis-NIRS) calibration models for predicting the content of organic carbon (OC), total nitrogen (N), clay + silt, and pH values in soils from Argentina, employing different mathematical preprocessing techniques for spectral data. A total of 154 soil samples with contrasting physicochemical characteristics was selected, dried, and sieved to 2 mm prior to the analysis of OC, N, pH, and clay + silt using reference methods. Subsequently, the Vis-NIR spectrum (400 to 2500 nm) of each sample was obtained in reflectance mode. The sample set was randomly divided into two groups: one for calibration model development (80%) and the other for model validation (20%). Eight preprocessing techniques for spectral information were employed, and the best one for each parameter studied was selected based on the criteria of achieving the minimum standard error of cross-validation (EECV), maximum coefficient of determination of cross-validation (R2cv), and maximum residual predictive deviation (RPD). Calibration models were obtained using modified partial least squares regression (MPLS) and five-fold cross-validation. The best models for predicting OC and N were obtained with the raw spectra (RPD=4.69 and 3.65 respectively); for pH, the best model was obtained using the standard normal variate preprocessing technique with the second derivative (RPD=2.27), and for clay + silt, the best model was obtained with multiple scattering correction and second derivative (RPD=2.83). The performance of the models applied to the calibration and testing groups was similar. The results indicate that the appropriate choice of spectral data preprocessing technique can optimize calibration for the prediction. Vis-NIRS is a valuable tool for soil monitoring in Argentina, complementing traditional analysis methods.

RESUMEN El objetivo de este estudio fue desarrollar modelos de calibración por espectroscopía visible e infrarrojo cercano (Vis-NIRS) para la predicción del contenido de carbono orgánico (CO), nitrógeno total (N), arcilla+li-mo, y pH en suelos de Argentina, aplicando diferentes técnicas de pretratamiento matemático de los datos espectrales. Se seleccionaron 154 muestras de suelo con características físico químicas contrastantes, se secaron y tamizaron a 2 mm previo al análisis de CO, N, pH y arcilla+limo mediante métodos de referencia. Luego se obtuvo el espectro Vis-NIR (400 a 2500 nm) de cada muestra en modo reflectancia. El conjunto de muestras fue dividido aleatoriamente en dos grupos: uno para obtener el modelo de calibración (80 %) y el otro para la validación del modelo (20 %). Se utilizaron ocho técnicas de pretratamiento de la información espectral y luego se seleccionó la mejor para cada parámetro, con el criterio de obtener el mínimo error estándar de la validación cruzada (EEVC), máximo coeficiente de determinación de la validación cruzada (R2cv) y máxima desviación predictiva residual (RPD). Los modelos de calibración se obtuvieron utilizando regresión de cuadrados mínimos parciales modificada (MPLS) y validación cruzada con cinco etapas. Los mejores modelos para la predicción de CO y N se obtuvieron con los espectros sin procesar (RPD=4,69 y 3,65 respectivamente); para pH se obtuvo con la técnica variación normal estándar con segunda derivada (RPD=2,27) y para arcilla+limo con corrección de dispersión múltiple y segunda derivada (RPD=2,83). El desempeño de los modelos aplicados sobre el grupo de testeo y calibración fue similar. Los resultados indican que la elección apropiada de la técnica de pretratamiento de datos espectrales puede optimizar la calibración. Vis-NIRS es una herramienta valiosa para el monitoreo de suelos en Argentina, como comple-mento de los métodos tradicionales de análisis.

ABSTRACT Rainfall forecasts are of great importance to prevent disasters, economic losses in crops and several important aspects for the country's hydroeconomy, hence the need to develop rainfall prediction models. At the Villa Clara Provincial Meteorological Center (CMPVC), mathematical models with a high level of accuracy are used, but they require a lot of processing time. Hence, the need to look for more current and accurate, ways to make accurate rainfall forecasts but in a faster way. The progress of artificial intelligence and especially artificial neural networks have allowed the development of models to solve the problem of rainfall prediction; using the metrics provided by the World Meteorological Organization (WMO) from 1975 to 2016. In this work, an analysis of the different regression models with multiple outputs to predict rainfall is carry out in the months of May, June, July and August. The regression models that best fit this problem were linear regression, regression tree, k-nearest neighbors, direct regression, chained regression and multilayer perceptron. After running the models, the multilayer perceptron was the regression model with the best results, with a high degree of efficiency for rainfall forecasting.

RESUMEN Los pronósticos de las precipitaciones tienen gran importancia para prevenir desastres, pérdidas económicas en cultivos y varios aspectos importantes para la hidroeconomía del país, de ahí la necesidad de desarrollar modelos de predicción de precipitaciones. En el Centro Meteorológico Provincial de Villa Clara (CMPVC) se emplean modelos matemáticos con un nivel de precisión alto, pero requieren mucho tiempo de procesamiento. De ahí la necesidad de buscar maneras más actuales y precisas para realizar pronósticos de lluvia acertados, pero de manera más rápida. El progreso de la inteligencia artificial y en especial, las redes neuronales artificiales han permitido desarrollar modelos para dar solución al problema de predicción de precipitaciones; mediante las métricas proporcionadas por la Organización Meteorológica Mundial (OMM) comprendidas entre los años 1975 hasta el 2016. En este trabajo se realiza un análisis de los diferentes modelos de regresión con múltiples salidas para predecir las precipitaciones en los meses mayo, junio, julio y agosto. Los modelos de regresión que más se ajustaron a esta problemática fueron la regresión lineal, el árbol de regresión, el k-vecinos más cercanos, la regresión directa, la regresión encadenada y el perceptrón multicapa. Luego de realizar las corridas de los modelos resultó el perceptrón multicapa, el modelo de regresión con mejores resultados, con un alto grado de eficacia para el pronóstico de las precipitaciones.