A machine learning framework for parkinson’s disease classification using digitized spiral drawings: markov-based feature selection and ensemble evaluation

Parkinson’s disease (PD) is a neurodegenerative disorder associated with progressive degeneration of dopaminergic neurons, resulting in motor and non-motor impairments. Digitized spiral drawings provide a non-invasive means of assessing motor abnormalities, but manual evaluation can be subjective and time-consuming. This study presents a machine-learning framework for classifying PD and healthy controls using features extracted from digitized spiral drawings. Six spatial and image-derived features are scaled using Min–Max normalization and ranked using a steady-state Markov chain feature selector, whose transition matrix combines absolute feature correlations with variability-based weights. Classification is performed using a stacking ensemble comprising a Decision Tree, a radial basis function support vector machine (RBF-SVM), and a Random Forest, with Logistic Regression as the meta-learner trained on out-of-fold predictions. Scaling, feature selection, and stacking are fitted within the training folds, while duplicate-linked files are assigned to the same split group. The stacking model had a five-fold cross-validation accuracy of 86.46 ± 17.17% and balanced accuracy of 84.23 ± 20.15% on the development set of 62 files. Sensitivity in out-of-fold pooling was 95.92%, while the specificity was 38.46%, implying that there were few healthy control identifications. However, on the hold-out set where all 15 files were classified, there were just two controls, and the 95% Wilson confidence interval for accuracy was 79.6–100%. Ablation experiments showed that stacking did not outperform a single RBF-SVM and that Markov-based feature selection did not outperform the use of all six features. The results prove the viability of the proposed method but do not indicate its diagnostic ability. More independently collected datasets would be required for assessing its generalizability.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-18
DOI
https://doi.org/10.1038/s41598-026-80953-9
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A machine learning framework for parkinson’s disease classification using digitized spiral drawings: markov-based feature selection and ensemble evaluation

Meenu Vijarania, Oluwasegun Julius Aroba, Joseph Bamidele Awotunde, Poonam Yadav et al.
Scientific Reports
Artificial Intelligence in Healthcare
article

A machine learning framework for parkinson’s disease classification using digitized spiral drawings: markov-based feature selection and ensemble evaluation

Meenu Vijarania, Oluwasegun Julius Aroba, Joseph Bamidele Awotunde, Poonam Yadav, Hyeladi Haruna
article en

Abstract

Parkinson’s disease (PD) is a neurodegenerative disorder associated with progressive degeneration of dopaminergic neurons, resulting in motor and non-motor impairments. Digitized spiral drawings provide a non-invasive means of assessing motor abnormalities, but manual evaluation can be subjective and time-consuming. This study presents a machine-learning framework for classifying PD and healthy controls using features extracted from digitized spiral drawings. Six spatial and image-derived features are scaled using Min–Max normalization and ranked using a steady-state Markov chain feature selector, whose transition matrix combines absolute feature correlations with variability-based weights. Classification is performed using a stacking ensemble comprising a Decision Tree, a radial basis function support vector machine (RBF-SVM), and a Random Forest, with Logistic Regression as the meta-learner trained on out-of-fold predictions. Scaling, feature selection, and stacking are fitted within the training folds, while duplicate-linked files are assigned to the same split group. The stacking model had a five-fold cross-validation accuracy of 86.46 ± 17.17% and balanced accuracy of 84.23 ± 20.15% on the development set of 62 files. Sensitivity in out-of-fold pooling was 95.92%, while the specificity was 38.46%, implying that there were few healthy control identifications. However, on the hold-out set where all 15 files were classified, there were just two controls, and the 95% Wilson confidence interval for accuracy was 79.6–100%. Ablation experiments showed that stacking did not outperform a single RBF-SVM and that Markov-based feature selection did not outperform the use of all six features. The results prove the viability of the proposed method but do not indicate its diagnostic ability. More independently collected datasets would be required for assessing its generalizability.

Scientific Reports
University of Ilorin (NG), Recep Tayyip Erdoğan University (TR), Durban University of Technology (ZA), KR Mangalam University (IN), Sharda University (IN)
Good health and well-being
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon