Multi Domain Anomaly Detection and Intent Prediction Engine

Inspiration

Modern security and defense environments increasingly operate in gray zones, where threats are ambiguous, low-confidence, and distributed across multiple domains. Commanders often receive hundreds of isolated alerts from physical sensors, cyber monitoring systems, and human intelligence sources. However, no single alert is strong enough to justify action. This situation leads to alert fatigue, delayed responses, and missed coordinated attacks.

We were inspired to build the Multi-Domain Anomaly Detection and Intent Prediction Engine (MDAD) to fuse weak signals across domains, reduce uncertainty, and support confident, proactive decision-making.

What it does

MDAD is an intelligence fusion and decision-support system designed to transform scattered low-confidence data into actionable insights. The system collects data from physical, cyber, and HUMINT domains and normalizes events across time, location, and confidence levels. It correlates signals using spatial, temporal, and semantic analysis, and applies Bayesian fusion to compute a unified threat score.

Based on detected patterns, MDAD predicts adversary intent and presents results through interactive maps, dashboards, and alert views. Instead of generating more alerts, MDAD reduces noise and highlights only meaningful, high-confidence threats for commanders.

How we built it

The MDAD system was designed as a modular pipeline consisting of the following components:

Data Ingestion Layer
Data Parsing and Normalization
Correlation Engine
Bayesian Fusion Engine
Intent Prediction Module
Visualization and Decision Support This modular architecture allows the system to be scalable, extensible, and adaptable to additional data sources and domains.

Challenges we ran into

One of the major challenges was handling heterogeneous data formats across multiple domains. Aligning events that were close in time but not perfectly synchronized required careful temporal normalization. Assigning meaningful confidence weights to HUMINT inputs was challenging due to their subjective nature.

We also had to prevent false positives when signals partially overlapped and ensure that probabilistic outputs were explained clearly to non-technical users. Each of these challenges required careful design choices to ensure the system remained accurate, interpretable, and usable.

Accomplishments that we’re proud of

We successfully fused multi-domain intelligence into a single unified pipeline and reduced hundreds of raw alerts into a small number of high-confidence threat indicators. The system goes beyond anomaly detection by incorporating intent prediction, enabling proactive rather than reactive responses.

Most importantly, MDAD demonstrates how weak and uncertain signals become powerful when they are fused correctly across domains.

What we learned

We learned that correlation across domains is significantly more powerful than isolated detection within a single domain. Bayesian approaches are highly effective for reasoning under uncertainty, especially in intelligence-driven environments. Visualization plays a critical role in decision-support systems and is just as important as the underlying detection algorithms.

We also learned that HUMINT data requires special handling due to subjective confidence levels, and that effective decision-support systems must prioritize clarity over complexity. This project deepened our understanding of multi-domain intelligence, probabilistic reasoning, and secure system design.