Skip to main content

​
Overview

The Behavior Analysis tool automatically classifies time series data into recognizable patterns using statistical model comparison. Whether you’re analyzing simulation outputs, historical data, or exploring synthetic patterns, this tool provides instant insights into the underlying behavior of your data.

​
Core Technology

​
Partnership with BEAMS

This behavior analysis capability was developed in partnership with the BEAMS (Benchmarking and Evaluating AI for Modeling and Simulation) project, which includes members of the system dynamics community and technical committee. The BEAMS project focuses on advancing AI applications in system dynamics and model-based analysis. For more details about the BEAMS project, visit: https://www.buffalo.edu/ai-data-science/research/beams.html This tool is part of the open source SD-AI project. The complete codebase and documentation can be found at: https://ub-iad.github.io/sd-ai/

​
Statistical Pattern Detection

The analyzer uses model selection to compare multiple behavioral models and identify the best fit:
  • Model Competition: Multiple pattern models compete to explain your data
  • Automatic Selection: Best-fit model chosen based on statistical likelihood
  • Confidence Scoring: Probability weights show detection confidence
  • Robust Normalization: Z-score normalization for scale-independent analysis

​
Detected Behavior Patterns

The analyzer can identify 12+ distinct behavioral patterns commonly found in system dynamics and time series data:

​
Growth & Decline Patterns

​
Cyclical Patterns

​
Transition Patterns

​
Stability Patterns

​
Using the Tool

​
Synthetic Data Mode

Generate and analyze preset behavior patterns: Synthetic Data Mode - Select patterns and view detected behaviors
  1. Select a Pattern: Choose from the pattern library dropdown
  2. Adjust Parameters:
    • Number of points (10-500)
    • Noise level (0-50%)
    • Amplitude, frequency, growth rate
  3. View Results: See instant classification with confidence scores
Try adding noise to see how robust the detection is! Most patterns remain detectable up to 20-30% noise.

​
Upload Your Data

Analyze your own CSV files: Upload Your Data - Drag and drop CSV files for analysis
  1. Drag and Drop: Drop a CSV file onto the upload area
  2. Format Requirements:
    • First row must contain column headers
    • Each column is analyzed as a separate variable
    • Optional: Include a “Time” or datetime column
  3. Multi-Variable Analysis: All numeric columns analyzed simultaneously

​
CSV Format Examples

Simple numeric data:
With datetime index:

​
Understanding Results

​
Confidence Interpretation

​
Scale Statistics

The analyzer provides comprehensive scale metrics:
  • Mean, Std, Min, Max: Basic distribution statistics
  • Range: Total variation in the data
  • Start/End Values: First and last observed values
  • Delta & Delta %: Absolute and percentage change
  • Direction: Overall trend (increasing/decreasing/stable)
  • R²: Linear fit quality (how well a line fits)
  • Correlation with Time: Trend strength

​
Handling Edge Cases

If the shape RMSE is high (above 0.85), the analyzer flags the result as “Possibly Complex or Unmodeled”. This indicates the data may have:
  • Multiple overlapping patterns
  • Non-standard behavior
  • High noise levels
  • Structural breaks
Data close to zero is automatically detected and classified as “Inactive” when appropriate, using absolute thresholds rather than relative measures.
Constant or near-constant data is detected using coefficient of variation (CV below 0.1) and classified as “Stable”.

​
Use Cases

​
Model Validation

Compare your simulation outputs against expected behaviors:
  • Verify exponential growth models produce exponential patterns
  • Confirm oscillating systems show proper dampening
  • Detect unexpected step changes or transitions

​
Data Exploration

Quickly characterize unfamiliar datasets:
  • Identify dominant patterns in historical data
  • Screen multiple variables for interesting behaviors
  • Find cyclical patterns in operational data

​
Teaching & Learning

Understand system dynamics archetypes:
  • Generate synthetic examples of each pattern
  • See how parameters affect behavior
  • Learn pattern recognition visually

​
Technical Details

​
Performance

  • Analysis Time: Typically under 1 second per variable
  • Memory: Runs entirely in browser memory
  • Data Limits: Optimized for series up to 10,000 points

​
Supported Formats

  • CSV files with numeric columns
  • Optional datetime column (auto-detected)
  • Missing values handled via linear interpolation
The behavior analysis tool uses the same Pyodide runtime as the simulation engine, so if you’ve run a simulation, the Python environment is already loaded.