Robust Feature Engineering for Parkinson Disease Diagnosis: New Machine Learning Techniques
Bibliographic record
Abstract
Background Parkinson disease (PD) is a common neurodegenerative disorder that affects between 7 and 10 million people worldwide. No objective test for PD currently exists, and studies suggest misdiagnosis rates of up to 34%. Machine learning (ML) presents an opportunity to improve diagnosis; however, the size and nature of data sets make it difficult to generalize the performance of ML models to real-world applications. Objective This study aims to consolidate prior work and introduce new techniques in feature engineering and ML for diagnosis based on vowel phonation. Additional features and ML techniques were introduced, showing major performance improvements on the large mPower vocal phonation data set. Methods We used 1600 randomly selected /aa/ phonation samples from the entire data set to derive rules for filtering out faulty samples from the data set. The application of these rules, along with a joint age-gender balancing filter, results in a data set of 511 PD patients and 511 controls. We calculated features on a 1.5-second window of audio, beginning at the 1-second mark, for a support vector machine. This was evaluated with 10-fold cross-validation (CV), with stratification for balancing the number of patients and controls for each CV fold. Results We showed that the features used in prior literature do not perform well when extrapolated to the much larger mPower data set. Owing to the natural variation in speech, the separation of patients and controls is not as simple as previously believed. We presented significant performance improvements using additional novel features (with 88.6% certainty, derived from a Bayesian correlated t test) in separating patients and controls, with accuracy exceeding 58%. Conclusions The results are promising, showing the potential for ML in detecting symptoms imperceptible to a neurologist.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".