Time-Series & Sensors

Reliable FRDA Assessment with Isolation Forest Anomaly Detection

Reliable Objective Assessment of Friedreich Ataxia Through Isolation Forest-Based Anomaly Detection

Automatic quality control for sensor recordings: unusual segments are found and set aside before they distort the score.

The problem

Irregular task execution or outside disturbances leave anomalies in the recordings that undermine severity predictions.

The approach

An Isolation Forest isolates anomalous segments, and adaptive median filtering guided by anomaly scores cleans the remaining signal.

The outcome

More reliable kinematic and kinetic features and higher test-retest agreement (ICC) for severity predictions.

In detail

Designed an anomaly-detection pipeline for AIM-C sensor data used in objective Friedreich ataxia assessment. Isolation Forest identifies atypical task executions caused by irregular movement or external disturbance, while adaptive filtering improves signal quality before feature extraction. The work focuses on improving the repeatability and reliability of sensor-derived severity predictions.

Built with

Source

More in Time-Series & Sensors

Next project: Objective FRDA Assessment in Children

By Kanishka Ranaweera, PhD, Machine Learning Engineer in Melbourne, Victoria, Australia. hello@kanishkabuilt.me