Parkinson Detection

ML on physiological time series for early diagnosis

CASE STUDY · HEALTH-DATA RESEARCH PROJECT

MODALITY

Motor signals

FORMAT

Per-subject TXT

APPROACH

Classical ML

NEXT STEP

Deep learning

STACK //scikit-learnSignal processingFeature engineering

STEP / 01

Brief

Parkinson leaves a fingerprint in motor signals long before clinical signs are obvious. The task: detect that fingerprint from raw physiological recordings.

STEP / 02

Pipeline

From per-subject text files to discriminative features.

01Per-subject text files parsed into clean time series
02Aggressive preprocessing — denoising, alignment, normalisation
03Hand-crafted features capturing tremor, rigidity, amplitude
04Model comparison and separability analysis

STEP / 03

Model

Classical classifiers benchmarked against each other to establish a baseline. The architecture is intentionally interpretable — clinicians need to see the lever the model is pulling.

STEP / 04

Friction points

Health data is rarely clean.

01Inter-subject variability dominates the signal
02Noisy acquisitions, missing windows
03Limited cohort size — overfitting risk is real

STEP / 05

Outcome

Discriminative patterns surfaced, reproducible pipeline, and a clean foundation for a deep-learning iteration on a larger cohort.

ML applied where the cost of being wrong is measured in years of someone's life.

NEXT CASE / P01

CAN Anomaly Detection

←BACK TO INDEXEND / P04