Plot Joint Probability

This demo demonstrates using joint probability analysis to detect bad channels based on multi-dimensional statistical outliers.

What is Joint Probability?

Joint probability analysis detects outlier channels by examining how each channel's distribution relates to all other channels simultaneously:

• Multi-dimensional comparison: Unlike single metrics (variance, kurtosis), joint probability considers the full distribution

• Statistical outlier detection: Channels that don't fit the overall pattern are flagged

• More sensitive: Catches unusual channels that may pass single-metric tests

How It Works

1. Calculate channel statistics: Extract features from each channel (amplitude, variance, etc.)

2. Build joint distribution: Model the typical pattern across all channels

3. Compute probabilities: For each channel, calculate how likely its pattern is

4. Identify outliers: Channels with very low probability are suspicious

Interpreting Results

Low probability values (close to 0):

• Channel doesn't fit typical pattern

• Potential bad channel

• May have unusual artifacts or noise

Consistent probability values (similar across channels):

• Data quality is uniform

• No obvious outliers

Spatial clusters of low probability:

• Systematic recording issues

• Poor contact in a region

• Regional artifacts

Use Cases

Pre-processing quality control:

• Automated bad channel detection

• Objective identification criteria

• Reduces manual inspection time

Complement other metrics:

• Use with variance and kurtosis

• Catches different types of artifacts

• More comprehensive quality assessment

Research validation:

• Document channel exclusion decisions

• Consistent criteria across datasets

• Reproducible preprocessing

Threshold Selection

The default threshold identifies statistical outliers, but you may need to adjust based on:

• Data quality: Noisy data may need more lenient thresholds

• Channel count: More channels → more statistical power

• Analysis goals: Conservative vs aggressive cleaning

Advantages Over Single Metrics

Variance alone: Misses channels with unusual distributions but normal variance

Kurtosis alone: Misses channels with shifted means or different scales

Joint probability: Combines multiple dimensions to catch subtle issues

Workflow Summary

This demo shows joint probability analysis:

Load and Preprocess

• Read continuous data

• Apply basic preprocessing (reference, filter)

Compute Joint Probability

• Calculate multi-dimensional statistics

• Build probability distribution

• Returns metrics for each channel

Visualize Results

• Plot probability values per channel

• Identify low-probability outliers

• Assess spatial patterns

Use Results for QC

• Flag suspicious channels

• Combine with other metrics

• Make informed exclusion decisions

Code Examples

Show Code

# Demo: Joint Probability Plot
# Shows joint probability distributions for artifact detection.

using EegFun
# Note: EegFun.example_path() resolves bundled example data paths.
# When using your own data, simply pass the file path directly, e.g.:
# dat = EegFun.read_raw_data("/path/to/your/data.bdf")

# read raw data
dat = EegFun.read_raw_data(EegFun.example_path("data/bdf/example1.bdf"));

# read and prepare layout file
layout = EegFun.read_layout(EegFun.example_path("layouts/biosemi/biosemi72.csv"));

# create EegFun data structure (EegFun.ContinuousData)
dat = EegFun.create_eegfun_data(dat, layout);

# Some minimal preprocessing (average reference and highpass filter)
EegFun.rereference!(dat, :avg)
EegFun.highpass_filter!(dat, 1)

# test plot_channel_summary
jp = EegFun.channel_joint_probability(dat)
EegFun.plot_joint_probability(jp)

See Also

• API Reference