Unidad de Posgrados

Permanent URI for this communityhttp://repositorio.uta.edu.ec/handle/123456789/11

Browse

Start date: 2022Subject: search.filters.subject.MACHINE LEARNING

Search Results

Now showing 1 - 4 of 4
  • No Thumbnail Available
    Item
    Aplicación de algoritmos de Machine Learning para predecir la deserción estudiantil en alumnos de primer y segundo semestre en universidades públicas del Ecuador.
    (Universidad Técnica de Ambato. Facultad de Ingeniería en Sistemas, Electrónica e Industrial. Maestría en Matemática Aplicada, 2023) Rodríguez Vásconez, Cristóbal Alejandro; Benalcázar Palacios, Marco Enrique
    Se estima que en Ecuador la tasa de deserción en los dos primeros semestres de universidad es del 20%. Existen factores socioeconómicos que influyen en el abandono académico de un estudiante. La carencia de programas que atiendan la insatisfacción estudiantil provoca que no se detecten problemas a tiempo y no se puedan aplicar acciones correctivas oportunamente. En este proyecto se aplican técnicas de Machine Learning para predecir la deserción estudiantil a partir de factores seleccionados: socioeconómicos, psicológicos, demográficos y académicos. Partimos de la recolección y tratamiento de datos y se usaron Redes Neuronales Artificiales para crear un modelo que clasifica a un estudiante entre desertor o a salvo de deserción. Se evalúan las métricas Acurracy, sensibilidad y especificidad para determinar qué tan eficiente es el modelo. El modelo final es capaz de clasificar estudiantes a salvo de deserción de forma correcta el 87% de las veces y logra clasificar a desertores de forma correcta el 60% de las veces.
  • No Thumbnail Available
    Item
    Evaluación con red neuronal del proceso de corte láser por CO 2 en materiales compuestos de fibra de cabuya
    (Universidad Técnica de Ambato. Facultad de Ingeniería en Sistemas, Electrónica e Industrial. Maestría en Física Aplicada, 2022) Almache Barahona, Verónica Carolina; Pérez Salinas, Cristian Fabián
    The application of Machine Learning today has allowed the development of learning models to solve problems in different fields of industry. This research work focused on relating neural networks (ANN) with the manufacture of composite materials (polyester matrix + fiber cabuya) and CO2 laser cutting machining. The objective is to develop a neural network to evaluate the application of machine learning to predict the surface finish characteristic of the material under study. The established cutting parameters were laser power and cutting speed. The surface finish characteristic to be evaluated was the surface roughness of the cut composite material. The sheet of the constructed composite material was subjected to CO2 laser cutting, which generated a set of 84 specimens. Experimental data was generated by measuring surface roughness through laboratory tests. The programming of the neural network was done with the Scikit-learn package. This is one of the most widely used open source libraries for machine learning available in Python. The results achieved by the prediction of the network based on the experimental data are related to the values predicted by the neural network model (ANN) and the performance of the network was evaluated using statistical metrics. The statistical results obtained were 0.946, 0.139 and 0.301 corresponding to the coefficient of determination (R2), the mean square error (MSE), and the mean absolute error (MAE) respectively. Therefore, it could be concluded that the performance of the developed neural network has a high validity and ability to predict surface roughness.
  • No Thumbnail Available
    Item
    Sistema electrónico de control y etiquetado de molduras para cuadros en tiempo-real mediante machine learning
    (Universidad Técnica de Ambato. Facultad de Ingeniería en Sistemas, Electrónica e Industrial. Maestría en Electrónica y Automatización, 2022) Chávez Pico, David Alejandro; Herrera Garzón, Marco Antonio
    In this work, a real-time system is presented that performs the detection of the different types of moldings made from the analysis of the image processing of their different surfaces, silhouettes and colors. The need to use this electronic system for the control and labeling of moldings for paintings in real time through machine learning is to reduce production times and store the number of moldings manufactured in a database in order to avoid downtime on the part of workers and thus increase productivity in the factory. In addition, it has the tools, software and hardware to be able to do it, in this case a device called NVIDIA's Jetson Nano will be used for image analysis with its respective camera, which allows artificial vision applications. On the other hand, this control system allows the moldings manufactured to be labeled with their specific characteristics by means of a QR code to make it functional and practical in the factory. This will benefit production since the amount of molding will increase, since costs will be reduced and there will be an increase in profits. Another important aspect is that with this system it is intended to have scalability for the future due to the fact that there are different branches where the moldings are transported and it will help to have a more exact control from the time the load leaves until the load arrives to avoid delays in counting as conventionally. It has been done in recent years, this data can be taken directly to the accounting department who are in charge of the amount of production that is carried out in the factory and the amount that is transported to the headquarters and each of the branches.
  • No Thumbnail Available
    Item
    Evaluación con red neuronal del proceso de desgaste abrasivo de placas de un material compuesto de látex con partículas de caucho reciclado
    (Universidad Técnica de Ambato. Facultad de Ingeniería en Sistemas, Electrónica e Industrial. Maestría en Físisca Aplicada, 2022-02) Muñoz Valverde, Pablo Rafael; Pérez Salinas, Cristian Fabián
    In this thesis, a Machine Learning approach was investigated in the field of manufacturing new materials for industry. In particular, artificial neural networks were used to predict the Taber wear index (TDI) of latex plates and recycled rubber particles. In recent years, the application of Artificial Intelligence and in particular Machine Learning to scientific disciplines has increased substantially. The purpose was to evaluate how machine learning works, in particular neural networks, and how it should be applied to make a prediction. The preliminary phase of the work was to create the experimentally obtained data set necessary for the secondary phase, which includes the analysis and modeling of neural networks. The generation of the data set involved the manufacture of the material and wear tests based on the ISO 9352 standard. In the context of the neural network, the Google TensorFlow software was used through the Python3 interface. The model developed allows to predict the IDT of the plate taking as independent variables; the volumetric percentage of material, the rotational speed, applied load and the number of cycles. The performance of the network will be evaluated through statistical tests such as the mean square error (MSE), the mean absolute error (MAE) and the coefficient of determination (R2).