Our proposed scheme yields pattern recognition systems with high generalisability.The scheme enables fair comparison between different pattern recognition systems.It sets up correct decision boundaries and avoids misclassifying training patterns.It gives more information about the identified pattern for root cause analysis.The pattern recognition accuracy increased when the proposed scheme was adopted. Most applications of machine learning (ML) algorithms to control chart pattern recognition (CCPR) have focused on pattern detection and identification, rather than obtaining more detailed information about the pattern, which is important for effective assignable cause analysis. If real control chart data is not available for training purposes, synthetic patterns must be generated. Furthermore, pattern recognition accuracies achieved by different CCPR systems are usually not comparable since these were developed with different training data. How to create a diverse range of patterns for designing CCPR systems that can be compared and that are able to recognise a greater variety of patterns is an issue that needs studying.This paper presents a scheme to generate training patterns that addresses this issue of diversity and comparability. The scheme also comprises change point detection and mean change categorisation methods that implement nonlinear models (NLMs) for estimating abnormal pattern parameters. The effect of this new pattern generation scheme on the accuracy of pattern recognition has been studied using two ML algorithms: Support Vector Machine (SVM) and Probabilistic Neural Network (PNN).With the proposed pattern generation scheme, the mean pattern recognition accuracy was increased by 6.90% and 8.42% when SVM and PNN were used, respectively.
1 December 2019 in Expert Systems with Applications
E-commerce has grown considerably in Latin America in recent years due to the COVID-19 pandemic. E-commerce users in English-speaking and Chinese-speaking countries have web-based tools to compare the prices of products offered by various retailers. The task of product comparison is known as entity matching in the data-science domain. This paper proposes the first entity-matching system for product comparison in Spanish-speaking e-commerce. Given the lack of uniformity of e-commerce sites in Mexico, we opted for a bimodal entity-matching system that uses the image and textual description of products from two of the largest e-commerce stores in Mexico. State-of-the-art techniques in natural language processing and machine learning were used to develop this research. The resulting system achieves F1 values of approximately 80%, representing a significant step towards consolidating a product-matching system in Spanish-speaking e-commerce.
Color-feature-based wildfire pixel segmentation has become a challenging task extensively addressed in various research studies. Rule-based models aim to identify fire pixels in a binary manner by determining whether the pixel intensity exceeds a specified threshold value. The authors determine the thresholds by analyzing diverse collections of images that contain wildfires. This has resulted in a lack of consensus on the thresholds determined by various researchers, even when the same color space is used during the examination process. Additionally, determining fire pixels in a binary manner complicates the handling of uncertainty and vagueness in color information. This research aims to enhance fire-pixel segmentation by integrating color-based rule models with a fuzzy set approach to effectively address uncertainty and vagueness. The proposed approach automatically learns the optimal set of fuzzy operators and rules for fire detection to construct a combined model. To address the limitations of combining binary class labels, this approach modifies the rule form proposed by various authors to obtain a fuzzy set of data, such as a grayscale fire map, instead of a crisp set of data, such as a binary fire map. In addition, our proposal uses a genetic algorithm approach to construct the best combination model. The final binary form of the fire map is calculated using the widely used Otsu method. The presented method is evaluated qualitatively and quantitatively in a well-accepted dataset designed for wildfire pixel segmentation tasks. The model obtained outperforms state-of-the-art rules and traditional strategies for combining binary labels in the F-measure and IoU metrics.
El estudio tiene como objetivo conocer la Intención Emprendedora (IE) por género de los alumnos de la Universidad Tecnológica el Retoño, para determinar qué tan efectivas están siendo las estrategias para incentivarla. El emprendimiento es considerado como fuente de desarrollo económico y social, actualmente el número de mujeres emprendedoras va en aumento, investigaciones han demostrado que los estudios universitarios predominan en las mujeres que deciden emprender. Es por esto que en el presente trabajo se estudia la intención emprendedora de los estudiantes universitarios en dos puntos del tiempo cuatrimestre septiembre-diciembre 2019 (propedéutico) y cuatrimestre mayo-agosto 2020, mediante la aplicación de un instrumento, centrando la atención en la intención emprendedora por género. Utilizando la prueba no paramétrica de Friedman para medidas repetidas y la aproximación de F de Fisher para observaciones multivariadas. Los resultados obtenidos mostraron que la IE de las mujeres es mayor que la de los hombres, tanto en el cuatrimestre septiembre-diciembre 2019 como en el cuatrimestre mayo-agosto 2020, así mismo la IE de las mujeres se mantuvo en el mismo nivel durante los 2 periodos estudiados, con lo cual se puede inferir que la percepción de las mujeres universitarias para emprender está cambiando y va en aumento.