Building Machine Learning Detection Models for Security Data
For detection engineers and security data scientists who need to build, evaluate, deploy and maintain ML detectors on real security telemetry, and know when a rule would do better.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
Most security ML projects fail for reasons that have little to do with model architecture: the labels are wrong, the evaluation ignores base rates, the features leak the answer, or the model cannot be explained to the analyst who receives its alert. This course is built around those failure points.
You will frame detection problems (supervised classification, anomaly detection, ranking) and decide when ML beats a Sigma rule and when it does not. You will build labeled datasets from incident history, red-team exercises and public resources such as the OTRF Security Datasets and the EMBER malware feature set, and normalize telemetry to schemas such as OCSF and the Elastic Common Schema. Hands-on modules cover feature engineering for process trees, command lines, DNS and URL strings, and authentication sequences; gradient-boosted trees, sequence models and text embeddings for command-line and script classification; and evaluation with time-based splits, precision at alert budget, and cost-weighted metrics. The final modules address adversarial evasion and poisoning of security models, explainability for analysts, deployment patterns into SIEM and streaming pipelines (including Microsoft Sentinel, Elastic and Splunk machine learning tooling), and drift monitoring and retraining. Code is in Python with scikit-learn, XGBoost and PyTorch.
What you will be able to do
Frame a security detection problem as classification, anomaly detection or ranking, and justify ML over a rule or vice versa
Build labeled datasets from incident history, red-team data and public datasets while avoiding leakage
Engineer features from process, command-line, network, DNS and authentication telemetry normalized to a common schema
Train and compare gradient-boosted, sequence and embedding-based models for security classification
Evaluate detectors with time-based splits, realistic base rates and precision at a fixed alert budget
Assess and harden a security model against evasion and training-data poisoning
Deploy a model into a SIEM or streaming pipeline with explanations analysts can act on
Monitor deployed detectors for drift and run a safe retraining and rollback process
Syllabus
6 modules · 19 lessons · about 22 hours
1
Module 1: Framing Detection Problems
Choose the right problem and the right tool before writing code.3 lessons
Rules, Anomalies, Classifiers and Rankers: Choosing the Approach
Base Rates, Alert Budgets and the Cost of an Error
Mapping Model Coverage to MITRE ATT&CK
2
Module 2: Data and Labels
Ground truth is the hardest part of security ML.3 lessons
Normalizing Telemetry with OCSF and the Elastic Common Schema
Labeling from Incidents, Red-Team Exercises and Public Datasets
Leakage, Duplication and Time-Based Splits
3
Module 3: Features and Models
Representations and model families that work on security data.4 lessons
Features from Process Trees and Command Lines
Features from DNS, URLs and Network Flows
Gradient-Boosted Trees as the Strong Baseline
Sequence Models and Embeddings for Scripts and Command Lines
4
Module 4: Evaluation That Predicts Production
Measure what the SOC will actually experience.3 lessons
Precision at Alert Budget and Cost-Weighted Metrics
Error Analysis with Analysts in the Loop
Comparing Against the Existing Rule Set
5
Module 5: Adversarial Robustness and Explainability
Attackers adapt to your model; analysts must trust its output.3 lessons
Evasion Attacks on Security Classifiers
Poisoning Risks in Feedback-Trained Detectors
Explanations Analysts Can Act On
6
Module 6: Deployment and Lifecycle
Ship the model and keep it healthy.3 lessons
Deploying into SIEM and Streaming Pipelines
Drift Monitoring, Retraining and Rollback
Detection-as-Code Practices for ML Models
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 25 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 60 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by scikit-learn, PyTorch, XGBoost, MITRE, OCSF, Microsoft, Elastic, or Splunk. Product and company names are trademarks of their respective owners.
Learn what generative AI has changed about scams, from cloned voices and deepfake video calls to fluent phishing, and build a short set of verification habits that work no matter how convincing the fake is. Free, about one hour, with a 10-question check at the end.
A plain-language introduction to AI in cyber defense: the difference between classic machine learning detection and LLM assistants, where each shows up across the security stack, how attackers are using AI, and how to judge a security AI claim. No coding or security background beyond IT basics is required.
Practical rules for using general-purpose and security-specific AI assistants in security work: classifying what can be shared, sanitizing logs and malware samples, verifying output before acting, handling untrusted content that may carry prompt injection, and keeping an auditable record. Ends with a one-page team policy you can adopt.