Predicting the price and trends of financial instruments is a major challenge in the financial industry, impacting investment decision-making efficiency for various stakeholders.Although numerous and effective artificial intelligence techniques have been applied to time series analysis, the prediction of exchange rate movements in the Forex market still necessitates parsimonious, interpretable, and accurate solutions.This paper presents a novel methodology for predicting the short-term directional movement of the euro-dollar exchange rate using market data, specifically by measuring price action. The proposed methodology prioritizes using market inflection points and the multidimensional nature of the differences between uptrends and downtrends to construct a linear discriminant function (LDA). The core of our methodology is our novel Linear Classifier Configurator (LCC) which includes stages for data preparation, feature selection, and detection of underlying structures. We validate the results and interpretations using the statistical power of parametric tests. The experiments use market data of the euro-dollar exchange rate in 15-minute and 1-week time frames. Additionally, we incorporate a collection of intraday winning trades provided by an algorithmic trading model applied between January 1999 and April 2023. The proposed LCC methodology achieves an out-of-sample classification accuracy of 98.77%, outperforming other methodologies based on sophisticated approaches such as Long Short-Term Memory (LSTM), Deep reinforcement learning (DRL), Wavelet analysis (WA), Sentiment analysis of textual content, Support Vector Machines (SVM), and Genetic Algorithms (GA). Furthermore, our methodology improves financial performance and reduces risk exposure in trading strategies, as well as it is useful in selecting variables and transferable to other financial assets.

Introducción: La tasa de abandono estudiantil en las instituciones de educación superior es un problema global que requiere soluciones personalizadas y adaptadas a cada institución. El presente artículo se realiza en la Universidad Politécnica Estatal del Carchi (UPEC) en Ecuador. Objetivo: Identificar los factores determinantes del abandono universitario mediante un modelo logístico que permita predecir si un estudiante puede abandonar la universidad de manera prematura. Métodos: Como técnica de modelado estadístico multivariante se utiliza un modelo de regresión logística LR. La muestra de estudiantes corresponde a 3821 registros de matrículas en la UPEC. Resultados: Los resultados evidencian que el modelo logístico propuesto presenta una precisión del 90%. Además, se identificó una relación inversa entre la deserción estudiantil, el número de miembros en el hogar y el nivel educativo del padre y la madre; por lo que se evidencia la necesidad de instaurar medidas de apoyo y prevención específicas para los estudiantes de hogares con menores niveles educativos y mayores cargas familiares. Conclusiones: El abandono estudiantil en la UPEC ocurre principalmente durante el primer nivel de carrera, determinando además que los factores relevantes que influyen en la deserción estudiantil en esta universidad son: la edad, el número de miembros en el hogar, el nivel de estudio y el promedio académico. Esta información permitirá a las autoridades universitarias plantear medidas preventivas para mejorar la permanencia y el éxito académico de sus estudiantes.

The field of educational research increasingly emphasizes predictive modeling of academic performance, focusing on identifying determinants of student success and crafting models to forecast future achievements. This investigation evaluates the efficacy of different feature selection techniques in predicting mathematics performance among Ecuadorian students, based on data from the 2021-2022 cycle of the Ser Estudiante test. Employing supervised logistic regression for classification, the study compares three feature selection methods: selection based on the highest k-scores, recursive feature elimination with cross-validation (RFECV), and recursive feature elimination (RFE). The assessment reveals that both the highest k-scores and RFECV methods are highly effective in isolating the most pertinent features for predicting mathematical prowess. These methods achieved prediction accuracies exceeding 90%, with k-scores attaining 96% and RFECV 92%. Furthermore, they demonstrated remarkable recall (94% and 97%, respectively) and F1-Score (96% and 91%, respectively). Additionally, both methods presented Receiver Operating Characteristic (ROC) curves with an area under the curve (AUC) of 99%, signifying superior discriminatory capability. The findings illuminate the critical role of judicious feature selection in enhancing the precision of predictive models for academic performance, particularly in mathematics. The results advocate for the application of these techniques in pinpointing key factors contributing to student success. This study not only contributes to the methodological discourse in educational data analysis but also provides practical insights for the Ecuadorian education system in leveraging data-driven approaches to enhance student outcomes.