Please use this identifier to cite or link to this item:
Title: Detector predictivo de conexiones fraudulentas
Author: Martín Tinaquero, David
Director: García Font, Víctor
Tutor: Hernández Jiménez, Enric
Others: Universitat Oberta de Catalunya
Keywords: machine learning
deep learning
computer security
Issue Date: 4-Jun-2018
Publisher: Universitat Oberta de Catalunya
Abstract: The main purpose of this work is the design and implementation of an efficient fraudulent predictive detector that meets the needs expressed by the client (Ancert). To carry out the project, the real needs to be covered have been studied in depth. After that, sufficient and quality network frames have been obtained to cover the modeling phases of the predictive classifier. Subsequently, systems have been investigated to structure the files, which facilitate their exploration and massive treatment. Once the data has been structured in the Cassandra NoSQL DB, an investigation has been made of the different methods of Machine Learning and its application areas, choosing the one that has been considered most suitable for the project. Once I have decided that the design of the detector will be made through deep neural networks (Deep Learning) with Python and the TensorFlow libraries, we have started with the predictive algorithm modeling, which includes the stages of design, training, validation, testing and saving of the model for its subsequent reuse. The accuracy obtained with the training data was ~ 99% and ~ 92% with the test data. All this, with a negligible percentage (~ 0.05%) of false positives, relevant to avoid the non-availability of the service to the legal users of the system. The predictive detector has been used to classify a set of unlabeled network frames and perform their subsequent insertion into the Cassandra database. The total time spent to complete the classification of ~ 300,000 frames and insert them into the database has been less than 3 'in all tests performed.
Language: Spanish
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
dmt88TFM062018_presentación.pptx671.92 kBMicrosoft Powerpoint XMLView/Open
dmt88TFM062018_producto.zip33.88 MBUnknownView/Open
dmt88TFM0618memoria.pdfMemoria del TFM1.52 MBAdobe PDFView/Open
dmt88TFM0618presentación.pdfPresentación del TFM454.68 kBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons