Ciencias Exactas y Ciencias de la Salud
Permanent URI for this collectionhttps://hdl.handle.net/11285/551039
Pertenecen a esta colección Tesis y Trabajos de grado de las Maestrías correspondientes a las Escuelas de Ingeniería y Ciencias así como a Medicina y Ciencias de la Salud.
Browse
Search Results
- Enhancing BGP security with MAD anomaly detection system and machine learning techniques(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024-12) Romo Chavero, María Andrea; Cantoral Ceballos, José Antonio; emipsanchez; Botero Vega, Juan Felipe; Navarro Barrón, Francisco Javier; School of Engineering and Sciences; Campus Monterrey; Pérez Díaz, Jesús ArturoAnomalies in the Border Gateway Protocol (BGP) represent a signicant vulnerability in the Internet’s infrastructure, as they can cause widespread disruptions, trafc misdirection, and even security breaches. Proactive detection of these anomalies is vital to preserving network stability and preventing potential cyberattacks. In response to this challenge, we present the Median Absolute Deviation (MAD) anomaly detection system, which combines traditional statistical methods with advanced machine learning (ML) techniques for more precise and dynamic detection. Our approach introduces a novel adaptive threshold mechanism, allowing the system to adjust based on the changing conditions of network trafc. This dynamic thresholding signif- icantly improves the accuracy, precision, and F1-score of anomaly detection compared to the previous xed-threshold version. Additionally, we integrate the MAD system with a diverse ML classiers, including Random Forest, XGBoost, LightGBM, CatBoost, and ExtraTrees to enhance the system’s ability to identify complex patterns that indicate unusual BGP behavior.We evaluate our detection system on well-documented BGP anomaly events, such as the Slammer worm, Nimda, Code Red 1 v2, the Moscow blackout, and the Telekom Malaysia misconguration. The results show that our system when combined with ML models achieves an overall accuracy and F1-score of 0.99, demonstrating its effectiveness across various anomaly types. By using both statistical and ML models, the system is able to capture irregularities that could signal security threats, offering a more comprehensive detection solution.This research highlights the importance of combining statistical anomaly detection with ML to obtain a balance between accuracy and computational efciency. The system’s low resource requirements and minimal pre-processing make it highly scalable, allowing it to be potentially deployed in real-time on large-scale networks.
- Anomaly detection in a cost-effective conveyor belt testing system(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024-09) Solórzano Souza, Ana Paula; Navarro Durán, David; mtyahinojosa, emipsanchez; Galluzzi Aguilera, Renato; Sotelo Molina, Carlos Gustavo; Navarro Durán, David; School of Engineering and Sciences; Campus Ciudad de MéxicoThe industrial manufacturing sector faces significant challenges regarding energy efficiency and operational sustainability, particularly in the management of motor driven systems such as conveyor belts. While traditional diagnostic methods effectively identify faults, a critical gap remains in closing the control loop to enable autonomous corrective actions, specifically for mechanical tension regulation y conveyor belts. This research addresses this limitation by designing, implementing, and validating a low-cost, small-scale prototype capable of soft real-time energy monitoring, anomaly detection, and automatic physical correction. The system utilizes an ESP32 microcontroller and an INA219 sensor to analyze voltage and current signals, employing unsupervised machine learning algorithms such as Isolation Forest (IF) and One-Class Support Vector Machine (OC-SVM). This serves to diagnose tension states defined by specific deflection thresholds. To execute the physical corrections, the platform integrates a custom built rack and pinion mechanism driven by a servomotor, which automatically regulates the belt tension. Methodologically, the study characterized optimal operational conditions at a 30% PWM duty cycle and applied a median filter that reduced signal variability. Four models were trained using exclusively optimal-state data under univariate and multivariate configurations. Experimental results demonstrate that electrical parameters serve as reliable indicators of mechanical tension in a small scale conveyor belt. The Univariate One-Class SVM model applied to voltage yielded the highest performance, achieving an F1-Score of 0.8624. Meanwhile, the Multivariate OC-SVM model demonstrated high reliability with a Precision of 90.51% and a Recall of 72.95%. Upon detection of anomalies (slippage or excessive tightness), the system successfully triggered autonomous adjustments via a rack and pinion mechanism. These findings validate the feasibility of using accessible monitoring to implement intelligent, self-correcting maintenance systems, offering a scalable solution to minimize downtime and energy waste in industrial environments.
- Harnessing machine learning for short-to-long range weather forecasting: a Monterrey case study(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024-05) Machado Guillén, Gustavo de Jesús; Cruz Duarte, Jorge Mario; mtyahinojosa, emimmayorquin; Filus, Katarzyna; Falcón, Jesús Guillermo; Ibarra, Gerardo; Departamento de Ciencias Computacionales; Campus Monterrey; Conant, Santiago EnriqueWeather forecasting is crucial in adapting and integrating renewable energy sources, particularly in regions with complex climatic conditions like Monterrey. This study aims to provide reliable weather prediction methodologies by evaluating the performance of various traditional Machine Learning models, including Random Forest Regressor (RFR), Gradient Boosting Regressor (GBR), Support Vector Regressor (SVR), and Recurrent Neural Networks (RNN) such as SimpleRNN, Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Cascade LSTM, Bidirectional RNNs, and a novel Convolutional LSTM/LSTM architecture that handles spatial and temporal data. The research employs a dataset of historical weather data from Automatic Weather Stations and Advanced Baseline Imager Level 2 GOES-16 products, including key weather features like air temperature, solar radiation, wind speed, relative humidity, and precipitation. The models were trained and evaluated across different predictive ranges by combining distinct sampling and model output sizes. This study’s findings underscore the effectiveness of the Cascade LSTM models, achieving a Mean Absolute Error of 1.6 °C for 72-hour air temperature predictions and 85.79 W/m2 for solar radiation forecasts. The ConvLSTM/LSTM model also significantly improves short-term predictions, particularly for solar radiation and humidity. The main contribution of this work is a comprehensive methodology that can be generalized to other regions and datasets, supporting the nationwide implementation of localized machine-learning forecasting models. This methodology includes steps for data collection, preprocessing, creation of lagged features, and model implementation, as well as applying distinct approaches to forecasting by using autoregressive and fixed window models. This framework enables the development of accurate, region-specific forecasting models, facilitating better weather prediction and planning nationwide.
- Estimating occupancy level in indoor spaces using infrared values and environmental variables(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024) Ovando Franco, Angelo Jean Carlo; Ceballos Cancino, Héctor Gibrán; mtyahinojosa, emimmayorquin; Dávila Delgado, Juan Manuel; Minero Re, Erik Molino; School of engineering and Sciences; Campus Monterrey; Alvarado Uribe, JoannaImproving energy efficiency in indoor spaces is critical to reduce harmful effects of excessive energy consumption worldwide. For this reason, estimating occupancy level of people in indoor spaces has been identified as a significant contributor to improve energy efficiency and space utilization. In this thesis, in order to contribute to the solution of this problem, it is proposed to estimate occupancy level of people in enclosed spaces through an indirect approach based on environmental and infrared data, using Machine Learning (ML) techniques. The selected environmental variables are temperature, relative humidity, and atmospheric pressure. In the process, the values of five different workstations from a collaborative work area at Tecnologico de Monterrey were collected to determine the occupancy level of each workstation. To estimate occupancy, supervised ML algorithms were used, obtaining an average accuracy for each workstation of 93%, by using both environmental and infrared data, compared to ground truth counts during occupied hours. Our results show that infrared data plus environmental variables are more accurate than infrared-only sensors for estimating indoor occupancy. At the same way, Random Forest (RF) was the algorithm that reached the highest accuracy among Support Vector Machine (SVM), K-Nearest Neighbors (KNN).
- Deep Learning Approach for Alzheimer’s Disease Classification: Integrating Multimodal MRI and FDG- PET Imaging Through Dual Feature Extractors and Shared Neural Network Processing(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2024) Vega Guzmán, Sergio Eduardo; Alfaro Ponce, Mariel; emimmayorquin; Ochoa Ruíz, Gilberto; Chairez Oria, Jorge Isaac; Hernandez Sanchez, Alejandra; School of Engineering and Sciences; Campus Monterrey; Ramírez Nava, Gerardo JuliánAlzheimer’s disease (AD) is a progressive neurodegenerative disorder whose incidence is expected to grow in the coming years. Traditional diagnostic methods, such as MRI and FDG-PET, each provide valuable but limited insights into the disease’s pathology. This thesis researches the potential of a multimodal deep learning classifier to improve the diagnostic accuracy of AD by integrating MRI and FDG-PET imaging data in comparison to single modality implementations. The study proposes a lightweight neural architecture that uses the strengths of both imaging modalities, aiming to reduce computational costs while maintaining state-of-the-art diagnostic performance. The proposed model utilizes two pre-trained feature extractors, one for each imaging modality, fine-tuned to capture the relevant features from the dataset. The outputs of these extractors are fused into a single vector to form an enriched feature map that better describes the brain. Experimental results demonstrate that the multimodal classifier outperforms single modality classifiers, achieving an overall accuracy of 90% on the test dataset. The VGG19 model was the best feature extractor for both MRI and PET data since it showed superior performance when compared to the other experimental models, with an accuracy of 71.9% for MRI and 80.3% for PET images. The multimodal implementation also exhibited higher precision, recall, and F1 scores than the single-modality implementations. For instance, it achieved a precision of 0.90, recall of 0.94, and F1-score of 0.92 for the AD class and a precision of 0.89, recall of 0.82, and F1-score of 0.86 for the CN class. Furthermore, explainable AI techniques provided insights into the model’s decisionmaking process, revealing that it effectively utilizes both structural and metabolic information to distinguish between AD and cognitively normal (CN) subjects. This research adds supporting evidence into the potential of multimodal imaging and machine learning to enhance early detection and diagnosis of Alzheimer’s disease, offering a cost-effective solution suitable for widespread clinical applications.
- The identification of DoS and DDoS attacks to IoT devices in software defined networks by using machine learning and deep learning models(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2022-05) Almaraz Rivera, Josué Genaro; PEREZ DIAZ, JESUS ARTURO; 31169; Pérez Díaz, Jesús Arturo; puelquio/mscuervo; Trejo Rodríguez, Luis Ángel; Botero Vega, Juan Felipe; School of Engineering and Sciences; Campus Monterrey; Cantoral Ceballos, José AntonioThis thesis project explores and improves the current state of the art about detection techniques for Distributed Denial of Service (DDoS) attacks to Internet of Things (IoT) devices in Software Defined Networks (SDN), which as far as is known, is a big problem that network providers and data centers are still facing. Our planned solution for this problem started with the selection of strong Machine Learning (ML) and Deep Learning (DL) models from the current literature (such as Decision Trees and Recurrent Neural Networks), and their further evaluation under three feature sets from our balanced version of the Bot-IoT dataset, in order to evaluate the effects of different variables and avoid the dependencies produced by the Argus flow data generator. With this evaluation we achieved an average accuracy greater than 99% for binary and multiclass classifications, leveraging the categories and subcategories present in the Bot-IoT dataset, for the detection and identification of DDoS attacks based on Transport (UDP, TCP) and Application layer (HTTP) protocols. To extend the capacity of this Intrusion Detection System (IDS) we did a research stay in Colombia, with Universidad de Antioquia and in collaboration with Aligo (a cybersecurity company from Medellín). There, we created a new dataset based on real normal and attack traffic to physical IoT devices: the LATAM-DDoS-IoT dataset. We conducted binary and multiclass classifications with the DoS and the DDoS versions of this new dataset, getting an average accuracy of 99.967% and 98.872%, respectively. Then, we did two additional experiments combining our balanced version of the Bot-IoT dataset, applying transfer learning and a datasets concatenation, showing the differences between both domains and the generalization level we accomplished. Finally, we deployed our extended IDS (as a functional app built in Java and connected to an own cloud-hosted Python REST API) into a real-time SDN simulated environment, based on the Open Network Operating System (ONOS) controller and Mininet. We got a best accuracy of 94.608%, where 100% of the flows identified as attackers were correctly classified, and 91.406% of the attack flows were detected. This app can be further enhanced with the creation of an Intrusion Prevention System (IPS) as mitigation management strategy to stop the identified attackers.
- An ensemble forecasting framework for time series(Instituto Tecnológico y de Estudios Superiores de Monterrey, 2021-11) Saldaña Rodríguez, Alejandro; REGIS HERNANDEZ, FABIOLA; 331834; Espinoza García, Juan Carlos; emipsanchez; Regis Hernández, Fabiola; Murrieta Cortés, Beatriz; Escuela de Ingenieria y Ciencias; Campus QuerétaroForecasting for businesses is essential and, because small to medium sized enterprises cannot afford to spend the resources on accurate forecasting, the necessity to build step-by-step procedures that aid in this process is vital. Forecasting using machine learning or more complicated models comes with its own sets of challenges as many of them have parameters that are not directly interpreted to the variables. Ensemble Forecasting is a mixture between machine learning and forecasting and it uses many proven mathematical concepts such as the law of large numbers, the Jury theorem, and proven empirical evidence of these models outperforming the single models counterparts. This thesis proposes a new methodology to modernize and include the data analytics part of the cross industry standard process for data mining described in (CRISP-DM) to the time series analysis methodology proposed by George E. Box. The ensemble methods composed of linear combinations and majority-rule voting made better predictions and the new Ensemble Forecast model proposed in this thesis proved to be more accurate and precise than any other model including the other ensembling methods.