Course Grade Received: A
———————————————————
Applied Machine Learning in Cybersecurity
This course applies machine learning to real cybersecurity problems, moving from foundational AI/ML concepts through supervised, unsupervised, and deep learning models. Coursework covers spam and malware classification, intrusion detection with ensemble methods, anomaly detection on network traffic, and time-series analysis using LSTM networks. Hands-on projects use real-world datasets in Python, with emphasis on model evaluation, performance tuning, and the ethical and privacy considerations that govern AI-driven security tooling.
🧠 What This Course Demonstrates
This coursework demonstrates the ability to take a security problem, select an appropriate model, train and evaluate it against real data, and interpret the results in operational terms. It bridges the gap between security operations and data science: understanding not just which model to apply, but what its false-positive rate costs a SOC team, why a black-box model creates accountability risk, and where automated detection should defer to human review.
Key Skills & Topics
🤖 Models & Methods
- Supervised classification (k-NN, decision trees, Naive Bayes, logistic regression)
- Ensemble methods (random forests, boosting)
- Unsupervised learning and clustering for anomaly detection
- Neural networks, LSTM, and transfer learning
- Adversarial ML and generative models
📊 Applied Data Science
- Feature engineering, preprocessing, and scaling
- Model evaluation (precision, recall, F1, AUC-ROC, confusion matrix)
- Hyperparameter tuning and cross-validation
- Python stack: scikit-learn, NumPy, pandas, PyTorch, JupyterLab
🔐 Security Applications & Governance
- Spam/phishing, malware, and intrusion detection
- Time-series anomaly detection on traffic and system logs
- Model interpretability and accountability in security decisions
- Privacy, data protection (GDPR, CCPA), and algorithmic bias
Course Materials
Mathematical Foundations of ML Security Models
📐 An analysis of how linear algebra and probability distributions operate inside four machine learning models used in security contexts: Naive Bayes, logistic and softmax regression, neural networks, and Gaussian anomaly detection. Examines tensor representation of feature data, matrix operations underlying model computation, and how Bernoulli and Gaussian distributions shape classification behavior and detection thresholds.
Mod_1_Assignment_Fall2026_CYSE-420_fdMachine Learning for DDoS Detection
🌐 An assessment of how machine learning strengthens cybersecurity defenses, focused on distributed denial-of-service detection. Covers how models identify volumetric and behavioral traffic anomalies that static rule-based systems miss, alongside the practical limitations of deployment: data quality, adversarial evasion, and false-positive cost at scale.
Mod_1_DB_Fall2026_CYSE420Module 2 · Spam Detection with Logistic Regression and Naive Bayes
📧 An end-to-end SMS spam classification pipeline comparing logistic regression and multinomial Naive Bayes across two train/test splits. Evaluated with confusion matrices and ROC curves, after correcting a data leakage bug in the text vectorizer.
CYSE420_Assignment02_ReportModule 3 · Malware Detection from PE Headers
🦠 Static malware detection on 16,631 Windows executables using PCA, k-nearest neighbors, and decision trees. A leaked filename column and 2,980 duplicate headers had to be removed before the results meant anything.
CYSE420_Assignment03_final_report_09102026Module 4 · Gradient Boosting for Intrusion Detection
🌲 Gradient Boosting on NSL-KDD network traffic, tuned with five-fold cross-validation and compared with Random Forest and AdaBoost. Tuning moved ROC-AUC by only 0.0002, and the 19 missed intrusions turned out to be the real finding.
CYSE420_Assignment04_ReportModule 5 · Unsupervised Anomaly Detection with K-Means and DBSCAN
🔎 Malware detection by clustering, with no labels used during fitting. When malware made up 70 percent of the data, the anomaly assumption broke. At 5 percent, DBSCAN reached 0.910 ROC-AUC.
CYSE420_Assignment05_Report_fd-1Module 6 · Neural Network Intrusion Detection System
🧠A TensorFlow/Keras multi-layer perceptron tuned on a separate validation set and compared with the Module 4 ensembles on identical test data. It caught the same intrusions but raised six times the false alarms.
CYSE420_Assignment06_ReportModule 7 · Transfer Learning for LSTM Anomaly Detection
🔁 An LSTM pre-trained on TCP traffic and fine-tuned on scarce UDP/ICMP data, tested against an identical model trained from scratch. Accuracy was equivalent, but transfer took 5.7 times longer.
CYSE420_Assignment07_ReportModule 8 · Generative AI in Cybersecurity
🤖 How GANs, VAEs, and Transformers support security work through synthetic attack data, anomaly detection, and adversarial testing. Also covers the dual-use, privacy, and bias risks that come with them.
CYSE420_Assignment08_Paper