Machine Learning for Early Detection and Severity Classification in People with Parkinson's Disease

Machine Learning for Early Detection and Severity Classification in People with Parkinson's Disease | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Machine Learning for Early Detection and Severity Classification in People with Parkinson's Disease Juseon Hwang, Changhong Youm, Hwayoung Park, Bohyun Kim, Hyejin Choi, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5195774/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Early detection of Parkinson's disease (PD) and accurate assessment of disease progression are critical for optimizing treatment and rehabilitation. However, there is no consensus on how to effectively detect early-stage PD and classify motor symptom severity using gait analysis. This study evaluated the accuracy of machine learning (ML) models in classifying early and moderate stages of PD based on spatiotemporal gait features at different walking speeds. A total of 178 participants were recruited, including 103 individuals with PD (61 early-stage, 42 moderate-stage) and 75 healthy controls. Participants performed a walking test on a 24-meter walkway at three speeds: preferred walking speed (PWS), 20% faster (HWS), and 20% slower (LWS). Key features—walking speed at PWS, stride length at HWS, and the coefficient of variation (CV) of stride length at LWS—achieved a classification accuracy of 78.1% using the random forest algorithm. For early PD detection, stride length at HWS and CV at LWS provided 67.3% accuracy with Naïve Bayes. Walking speed at PWS was the most critical feature for distinguishing early from moderate PD, with an accuracy of 69.8%. These findings suggest that assessing gait over consecutive steps under different speed conditions may improve early detection and severity assessment of people of PD. Health sciences/Biomarkers Health sciences/Diseases Health sciences/Health care Parkinson’s disease gait severity motor symptom artificial intelligence machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 INTRODUCTION Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by the loss of dopaminergic neurons in the substantia nigra of the midbrain, which disrupts the function of the basal ganglia circuitry and impairs motor control [ 1 , 2 ]. One of the key motor symptoms of PD is the reduced ability to perform automated movements, such as walking [ 3 ]. Consequently, individuals with PD often experience significant locomotor difficulties, leading to a loss of independence and reduced quality of life (QoL) [ 4 ]. Given these challenges, early management of PD is crucial. Timely intervention may help delay disease progression, alleviate both motor and non-motor symptoms, and enhance the QoL of people with PD, potentially resulting in substantial cost savings for healthcare systems [ 5 ]. However, one of the primary obstacles to early intervention is that PD is often diagnosed at an advanced stage [ 5 ]. Additionally, distinguishing early-stage PD from normal age-related gait impairments is challenging, as both conditions may present with similar symptoms, such as decreased mobility and impaired balance, due to age-related degeneration in musculoskeletal and neurological functions [ 6 ]. Moreover, the early symptoms of PD are often subtle, leading to delayed diagnosis and treatment until the disease has progressed and symptoms become more pronounced and difficult to manage [ 7 ]. As a result, there is growing interest in using gait assessments as a potential diagnostic tool for detecting PD in its early stages. While gait assessments show promise for the early diagnosis of PD [ 8 – 10 ], further research is needed to refine these methods and improve their accuracy in distinguishing PD from other age-related conditions. Such advancements could enable earlier interventions, improve patient outcomes, and slow disease progression. Recently, there has been limited research on the early detection of PD through gait assessment using machine learning (ML) and deep learning algorithms based on artificial intelligence [ 11 – 13 ]. For instance, Ferreira et al. conducted a gait analysis to classify individuals with early-stage PD (Hoehn and Yahr [H&Y] stages 1 and 2) and healthy controls. The study demonstrated that the support vector machine (SVM) algorithm achieved an accuracy of 76.9% when utilizing all 22 gait features, whereas the Naïve Bayes (NB) classifier reached a higher accuracy of 84.6% using four selected features [ 11 ]. However, the study had a small sample size and relied on averaged data collected from noncontinuous gait analysis. This method does not fully replicate natural walking performance, which limits the reproducibility and generalizability of the findings. As PD progresses, typical pathological symptoms such as bradykinesia, rigidity, and impaired motor automaticity become increasingly pronounced. This often necessitates higher dosages of antiparkinson medications [ 14 , 15 ]. However, escalating dosages can lead to motor complications, including motor response fluctuations and dyskinesias [ 15 ]. The decision to initiate levodopa treatment is primarily influenced by the severity and rate of progression of the disease [ 16 ]. Accurate classification of PD stage is crucial for optimizing therapeutic interventions, as different stages exhibit distinct motor and non-motor characteristics. Such classifications enable clinicians to tailor exercise interventions and medical treatments to the specific needs of patients at various stages of the disease [ 16 ]. Subtyping PD according to disease stage could facilitate the development of personalized treatment protocols, potentially reducing dependence on medications associated with significant side effects, such as dyskinesias and motor fluctuations [ 15 , 17 ]. In a related study, Zhao et al. assessed gait performance using only self-selected walking speed to detect the severity of PD (H&Y stages 2, 2.5, and 3) [ 18 ]. The results revealed that the ensemble K-nearest neighbor algorithm achieved an accuracy of 76.03%. However, the study faced similar limitations, and since there remains no consensus on using gait variables to detect PD severity, further research is necessary to explore gait variables that can classify moderate stages of PD. Additionally, Additionally, Lee et al. recommended assessing gait ability across various walking speeds commonly encountered in daily life to detect the severity of motor symptoms in individuals with PD [ 19 ]. These more challenging walking tasks may provide a more accurate assessment of gait variability, as they demand greater executive function and attention to maintain dynamic stability [ 20 ]. Therefore, it is essential to consider challenging walking tasks under different speed conditions to identify markers of disease progression in individuals with PD. This study aimed to identify the accuracy of ML algorithms in classifying gait variables based on continuous gait tasks using three walking speeds, specifically to classify: i) people with PD versus age-matched healthy controls, ii) people with early PD (H&Y stage 1 to 2) versus age-matched healthy controls, and iii) people with moderate PD (H&Y stage 2.5 to 3) versus people with early PD (H&Y stage 1 to 2). METHODS Participants A total of 103 individuals with PD and 75 age-matched healthy controls participated in this study. The inclusion criteria for participants with PD were: a) a clinician-confirmed diagnosis of idiopathic PD according to the UK Parkinson’s Disease Society Brain Bank Clinical Diagnostic Criteria [21], b) age over 50 years, c) a modified H&Y scale score of 1–3, indicating independent walking and mobility without the need for an assistive device, and d) a score of ≥ 24 on the Korean Mini-Mental State Examination [22]. The control group consisted of age-matched healthy adults over 50 years old with no history of musculoskeletal, cardiovascular, vestibular, or neurological disorders that could have affected gait in the past six months (Figure 1). The physical and clinical characteristics of all participants are presented in Table 1. Group 1 consisted of 75 age-matched healthy controls. Participants with PD were further categorized based on their H&Y scale: Group 2 included 61 those in stages 1 to 2, and Group 3 included 42 those in stages 2.5 to 3. This study was approved by the Institutional Review Board of Dong-A University Hospital (IRB number: DAUHIRB-22-089). All participants were informed of the experimental procedures and provided written informed consent, as approved by the committee. The study is registered in the Clinical Research Information Service of the Republic of Korea (KCT0009353). Test Procedures and Data Analysis Gait analysis in this study was conducted using shoe-type data loggers (Smart Balance® SB-1, JEIOS, Korea) equipped with an inertial measurement unit (IMU) embedded in the shoe outsole [23]. The IMU sensor captured triaxial acceleration (up to ±6 g) and angular velocity (up to ±500°/s) along three orthogonal axes. The measured data were transmitted via Bluetooth to a gait analysis system (DynaStab™, JEIOS, Korea). Participants with PD performed the gait test while in the "ON" state of antiparkinsonian medication for at least 2 h, reflecting their typical gait performance during daily activities. All participants were first assessed by a clinician for clinical characteristics using the Korean Mini-Mental State Examination, the H&Y scale, and the Unified PD Rating Scale. Physical characteristics, including height, weight, shoulder width, hip and waist circumference, thigh and calf circumference, and leg length, were also measured. Participants began with a 5-minute warm-up involving stretching exercises before starting the experiment. Following the warm-up, participants performed the gait test while wearing shoes embedded with IMU sensors on a 24-meter straight walkway. Each participant was instructed to walk at a comfortable pace, with a metronome used to measure their self-selected walking speed in steps per minute (beats/min). Lower walking speed (LWS) and higher walking speed (HWS) were defined as 80% and 120% of the preferred walking speed (PWS), respectively [19]. Prior to the test, participants completed three practice trials. After a 5-minute break to wash out the effects of metronome-based pacing, participants performed the gait test at both HWS and LWS. All walking speeds were measured using the metronome (beats per minute), but the actual gait test was conducted without the metronome, as external auditory cues could affect the performance of individuals with PD [24] (Supplementary Figure 1). All gait test data were collected at 100 Hz and filtered using a second-order Butterworth low-pass filter with a cut-off frequency of 10 Hz. Spatiotemporal gait variables were analyzed over a 20-meter section of the walkway, excluding the 2-meter acceleration and deceleration phases at either end. The spatiotemporal gait variables were calculated as follows: gait speed (m/s), stride length (m), single-support phase (%), double-support phase (%), stance phase (%), the coefficient of variation (CV) for all spatiotemporal variables [25], and gait asymmetry [26]. Statistical Analysis All data were checked for normality using the Shapiro–Wilk test, and statistical analyses were performed using both parametric and nonparametric tests. The mean and standard deviation (SD) were calculated using one-way analysis of variance (ANOVA) for the physical characteristics of all participants, while independent t-tests and Mann–Whitney U tests were used to assess the clinical characteristics of individuals with PD. An intraclass correlation coefficient (ICC) analysis (2,1) was conducted to validate the agreement between the estimated and actual measured speeds under high- and low-speed conditions. The limits of agreement were calculated using a Bland–Altman plot to examine the error between the estimated and actual measured speeds [27]. To reduce the dimensionality of the input data, this study used the random forest (RF) method to assess feature importance. Specifically, the hyperparameters for the RF algorithms were tuned by experimenting with the number of trees set to 100, 200, and 300, and the maximum number of features set to 0.25, 0.5, and 0.75, while the random state was fixed at 42. This approach helps exclude the influence of irrelevant variables and facilitates data post-processing. We evaluated the importance of variables in three cases: (i) between age-matched healthy controls and individuals with PD, (ii) between age-matched healthy controls and individuals with early-stage PD, and (iii) between individuals with early-stage PD and those with moderate-stage PD. We ranked the variables in order of importance for each case and selected the top 15 spatiotemporal gait variables. Finally, we used multivariate logistic forward stepwise regression analysis to identify the optimal gait variables for classifying the three groups from these top 15 variables. Covariates included participants' age, gender, height, and body mass index. The cut-off values of the selected gait variables for group classification were examined using receiver operating characteristic (ROC) curves. The area under the ROC curve (AUC) was calculated to assess the ability of the selected variables to classify the groups [28]. Statistical significance was set at 0.05. Additionally, a ML approach was applied to explore the potential of spatiotemporal gait features for classifying individuals with PD versus age-matched controls, as well as for distinguishing between different PD severity levels. Several ML algorithms were employed to evaluate the suitability of these gait variables for classification across different groups, with each algorithm providing unique insights into the data, thus necessitating an assessment of variation in classification accuracy. To address the classification tasks, seven ML algorithms were applied (Figure 2): logistic regression (LR) [29], K-nearest neighbor (KNN) [30], NB [31], linear discriminant analysis (LDA) [32], quadratic discriminant analysis (QDA) [32], SVM [33], and random forest (RF) [34]. The model hyperparameters for each classifier were tuned using a grid search (Table 2). Five-fold cross-validation was used to evaluate the accuracy, precision, recall, and F1 scores for the classification performance of the seven algorithms. In this method, the dataset is divided into five parts, with four parts used as training data and one part used as test data in each iteration. The process is repeated five times, and the average of the results from the test sets is used as the final accuracy. Given the imbalanced dataset, which included 75 individuals in Group 1, 61 in Group 2, and 42 in Group 3, a random oversampling method was employed to address this imbalance [35]. Statistical analyses were performed using SPSS 21.0 (IBM Corp., Armonk, NY, USA) and Python (Python 3.10, Python Software Foundation). RESULTS Reliability of Estimated and Measured Speeds Supplementary Table S1 presents the reliability of the estimated and measured walking speeds at the LWS and HWS for both the control group and individuals with PD. The agreement between the estimated and measured walking speeds ranged from 92.0% to 95.1% at LWS and from 94.7% to 96.1% at HWS for both groups (Figure 3). Feature Selection for Group Classification The feature importance values for all 30 gait variables in each of the three classification cases are listed in Supplementary Table S2 (see Supplementary Tables S3, S4, and S5 for a comparison of gait variables across classification cases). The ranking of the importance of all variables, highlighting those ranked in the top 15 for each case (Figure 4, Supplementary Figure 2). Using these selected variables, the multivariate logistic forward stepwise regression analyses revealed significant differences between the control group and individuals with PD. Specifically, the CV for stride length at LWS (cut-off value: 2.64%; AUC: 0.409; sensitivity: 0.453; specificity: 0.456; odds ratio [OR]: 1.820), walking speed at PWS (cut-off value: 1.27 m/s; AUC: 0.781; sensitivity: 0.747; specificity: 0.738; OR: 0.421), and stride length at HWS (cut-off value: 1.43 m; AUC: 0.757; sensitivity: 0.680; specificity: 0.680; OR: 0.460) showed significant differences. In the classification between controls and individuals with early-stage PD, significant differences were observed in the CV of stride length at LWS (cut-off value: 2.62%; AUC: 0.405; sensitivity: 0.453; specificity: 0.459; OR: 1.707) and stride length at HWS (cut-off value: 1.48 m; AUC: 0.690; sensitivity: 0.613; specificity: 0.656; OR: 0.167). Additionally, in the classification between individuals with early-stage PD and those with moderate-stage PD, walking speed at PWS (cut-off value: 1.18 m/s; AUC: 0.739; sensitivity: 0.590; specificity: 0.738; OR: 0.376) showed significant differences (Table 3). ML Approach This study addressed three classification tasks using two distinct feature sets: one consisting of all spatiotemporal features, and the other comprising features selected through binary stepwise regression (Table 4, Figure 5). In the classification of controls and individuals with PD, the RF algorithm demonstrated the highest accuracy with all features (79.1% ± 5.4% SD). After feature reduction through stepwise regression, RF maintained the highest accuracy (78.1% ± 9.0% SD). For the early-stage classification of controls and individuals with PD, RF again achieved the highest accuracy (71.3% ± 6.9% SD) with all features, while NB achieved the highest accuracy (67.3% ± 8.6% SD) using two selected features. In classifying individuals with early-stage PD and moderate-stage PD, the SVM demonstrated the highest accuracy (75.5% ± 5.4% SD) with all features, while both QDA and SVM achieved the highest accuracy (69.8% ± 11.4% SD) with one selected feature. Additionally, classification performance metrics such as recall, precision, and F1 score were evaluated using a confusion matrix (Supplementary Table S6, Figure 6). DISCUSSION Classification of People with PD and Age-Matched Healthy Controls This study classified age-matched healthy controls and individuals with PD using gait variables under LWS, PWS, and HWS conditions. We demonstrated that individuals with PD exhibited a 57.9% reduction in walking speed at PWS and a 54.0% shortening of stride length at HWS compared to controls, which is consistent with previous reports [ 36 , 37 ]. Recently, Trabassi et al. evaluated the classification performance of ML models applied to individuals with PD and healthy controls, with the SVM achieving an accuracy of 81% for all 22 features and 86% for seven selected features, including the CV of step length, stride length, and stance phase [ 38 ]. In comparison, the present study achieved an accuracy of 79.1% for all 30 features and 78.1% for the three selected features (CV of stride length at LWS, walking speed at PWS, and stride length at HWS) using the RF algorithm, demonstrating similar performance to previous findings. Individuals with PD exhibit disrupted walking speeds, primarily due to difficulties in generating adequately sized steps and delays in movement timing, which are attributed to impaired cue production in the basal ganglia. This disruption may also involve the under-scaling of movement through the cortical-thalamic-basal ganglia circuitry [ 39 ]. However, none of the previous studies considered variable walking speeds or highlighted the importance of gait variability in their analyses. The result could be explained by the fact that walking at a preferred speed represents a gait pattern that minimizes variability and requires the least energy expenditure, regardless of cerebellar motor dysfunction or vestibular feedback control disorders [ 39 ]. In this study, we identified the variability of stride length at HWS as an important variable for distinguishing individuals with PD from controls. Increased variability in individuals with PD is thought to stem from a reduced ability to control gait timing, resulting from dysfunction in various components of the basal ganglia and other regions of the locomotor network. This dysfunction leads to a series of disjointed strides, which ultimately decrease walking speed [ 2 ]. Furthermore, both LWS and HWS are associated with prolonged single-limb stance phases and lateral displacement of the center of mass, further contributing to increased gait variability in individuals with PD [ 40 , 41 ]. Collectively, individuals with PD may experience difficulty walking under both slow and fast conditions, which can sensitively detect a decline in walking ability. Therefore, we suggest that evaluating gait under slower, faster, and preferred walking speeds may provide a valuable approach for overcoming these limitations [ 40 ]. Classification of Controls and People with Early-Stage PD As a primary finding, the classification between individuals with early-stage PD and age-matched healthy controls, using multivariate logistic forward stepwise regression analysis, identified two significant gait features: the CV of stride length at LWS (OR: 1.707; AUC: 0.405) and shortened stride length at HWS (OR: 0.167; AUC: 0.690), with an explanatory power of 40.4%. Additionally, the RF algorithm achieved an accuracy of 71.3% using all features, while NB reached an accuracy of 67.3% with the two selected features. In this study, individuals with early-stage PD demonstrated a 1.71-fold increase in the CV of stride length at LWS and an 83.3% reduction in stride length at HWS compared to age-matched healthy controls, consistent with previous studies [ 9 , 10 ]. Motor symptoms in early-stage PD often manifest as unilateral parkinsonism, primarily affecting one side of the body, due to asymmetrical degeneration of the basal ganglia [ 14 ]. Impaired basal ganglia function can result in reduced stride length during all gait phases due to bradykinesia and decreased walking velocity in individuals with early-stage PD [ 14 ]. A recent study by Ferreira et al. performed gait analysis on an 8.5 m walkway at self-selected walking speeds to classify individuals with early-stage PD (H&Y stages 1 and 2) [ 11 ]. The study found that SVM with a radial basis function kernel achieved an accuracy of 76.9% using all 22 features, while NB achieved an accuracy of 84.6% using four selected features, including walking speed, step length, step width, and the CV of step width. They suggested that the pace and variability domains of walking are the most discriminative features in classifying early-stage PD, which aligns with the findings of this study. In our study, the CV of step length at HWS and step length at LWS achieved a lower accuracy (67.3% using the NB algorithm) compared to the previous study, which used non-continuous gait analysis data. Non-continuous steps may better mimic natural walking performance but can limit reproducibility and generalizability [ 42 ]. Some researchers suggest that gait assessments on a 20 m walkway, with at least 30 steps, can improve the reliability and generalizability of the results [ 40 , 43 ]. Therefore, conducting gait assessments with at least 30 consecutive steps under various speed conditions may provide a more objective and sensitive method for classifying early-stage PD from age-matched healthy controls. Classification of Early and Moderate Stages of PD In the classification between individuals with early-stage PD and those with moderate-stage PD, the selected gait feature was walking speed at PWS (OR: 0.376; AUC: 0.739), with an explanatory power of 29.1%. Additionally, the SVM algorithm achieved an accuracy of 75.5% using all features, while both SVM and QDA achieved an accuracy of 69.8% using the single selected feature. As PD progresses, motor symptoms such as bradykinesia, rigidity, and bilateral involvement become more pronounced, reducing gait asymmetry compared to early-stage PD [ 14 ]. In the present study, walking speed at PWS was 1.707 times lower in individuals with moderate-stage PD compared to those with early-stage PD. This finding highlights the significant discriminative power of walking speed in classifying individuals with PD between early and moderate stages. Previous studies have also demonstrated that walking speed deteriorates significantly as PD progresses, consistent with the findings of this study [ 44 ]. This emphasizes the importance of slower movement as a key independent factor in walking ability during the early-to-moderate stages of PD [ 45 ]. Recent classification studies using ML models have not specifically investigated the accuracy of classifying individuals with PD across H&Y stages 1 to 3 based on spatiotemporal gait features from straight walkways. In this study, we achieved an accuracy of 69.8% for walking speed at PWS, a key feature in moderate-stage PD. Although this accuracy is slightly lower than that reported in previous studies [ 10 , 46 ], the use of sufficient consecutive steps allowed us to achieve reliable accuracy in classifying PD in the early-to-moderate stages [ 40 ]. Thus, gait assessments involving at least 30 consecutive steps under various speed conditions may provide a more objective and sensitive method for classifying individuals with PD across these stages. The present study offers valuable clinical insights. First, a marker based on three walking speeds with at least 30 consecutive steps was developed for H&Y stage classification. This marker may aid in distinguishing early-stage PD from normal aging and in classifying the severity of motor symptoms in PD. Furthermore, it has the potential to significantly impact clinical practice by facilitating personalized therapeutic and exercise interventions tailored to individual patient characteristics. Additionally, using selected gait features across three walking speeds could help physicians save time in assessing motor symptoms and provide a more objective evaluation of gait ability in clinical settings. This study has several limitations. First, we analyzed gait ability only during dopaminergic medication intake to reflect the daily activities of individuals with PD [ 46 , 47 ]. However, assessing gait performance in both the ON and OFF medication states would provide a more comprehensive evaluation, as both are crucial for understanding motor performance in PD. Second, we examined a limited dataset of individuals with early- and moderate-stage PD, which may have impacted the performance of the ML models [ 48 ]. Third, while the ML algorithms achieved promising results in classifying age-matched healthy controls and individuals with PD, the performance was slightly lower when classifying between early-stage and moderate-stage PD. To address these limitations, future studies should include multimodal datasets incorporating clinical, cognitive, motor, and neuroimaging variables, which may enhance the reliability and accuracy of characterizing PD progression [ 49 , 50 ]. CONCLUSION This study analyzed the accuracy of ML algorithms in classifying gait variables based on a continuous gait task performed at three different walking speeds, aimed at early detection and severity classification of PD. The key gait markers for distinguishing individuals with PD from healthy controls were walking speed at PWS, stride length at HWS, and the CV of stride length at LWS. These features achieved a classification accuracy of 78.1% using the RF algorithm. For early detection of PD, stride length at HWS and the CV of stride length at LWS demonstrated strong potential, achieving an accuracy of 67.3% with the NB algorithm. Furthermore, walking speed at PWS was the most effective feature for assessing disease severity, achieving an accuracy of 69.8% using SVM and QDA. These findings suggest that assessing gait over at least 30 consecutive steps at three different walking speeds could provide an objective and reliable method for evaluating motor symptoms in PD. Such assessments could serve as a valuable tool for clinicians and rehabilitation professionals, offering insights that support the development of personalized treatment and rehabilitation strategies tailored to the progression of the disease. Declarations ETHICS DECLARATIONS All study procedures involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study protocol and supplementary information files were approved by the Institutional Review Board of Dong-A University Hospital (IRB number: DAUHIRB-22-089) (see ethics approval letter in the supplementary file). All patients provided written informed consent before data collection. The study is registered with the Clinical Research Information Service in the Republic of Korea (KCT0009353). COMPETING INTERESTS The authors declare that they have no conflicts of interest. Consent for publication Not applicable. FUNDING This work was supported by a grant [Grant No. 2022R1A2C100933711; Changhong Youm] from the National Research Foundation of Korea (NRF), funded by the Korean government (MSIT). This research was also supported [Grant No. 2022R1A6A3A0108756411; Hwayoung Park] by the Basic Science Research Program through the NRF, funded by the Ministry of Education. The funders had no role in the study design, collection, analysis and interpretation of the data, and in writing the manuscript. Author Contribution JH, CY, HP, BK, HC, and SC conceived and designed the study. JH, HP, BK, HC, and SC recruited the participants. JH, CY, HP, BK, HC, and SC performed the data acquisition. JH, CY, HP, BK, HC, and SC analyzed and interpreted the data. JH, CY, HP, BK, HC, and SC drafted the article. All authors read and approved the final version of the manuscript submitted. Acknowledgement The authors thank all the participants of this study. This work was supported by a Dong-A University Research Fund. Data Availability The datasets supporting this study’s findings are available from the corresponding author upon reasonable request. References Alexander, G. E., DeLong, M. R. & Strick, P. L. Parallel organization of functionally segregated circuits linking basal ganglia and cortex. Annu. Rev. Neurosci. 9 , 357–381. https://doi.org/10.1146/annurev.ne.09.030186.002041 (1986). Takakusaki, K., Tomita, N. & Yano, M. Substrates for normal gait and pathophysiology of gait disturbances with respect to the basal ganglia dysfunction. J. Neurol. 255 , 19–29. https://doi.org/10.1007/s00415-008-4004-7 (2008). Smith, M. D., Brazier, D. E. & Henderson, E. J. Current perspectives on the assessment and management of gait disorders in Parkinson's disease. Neuropsychiatr Dis. Treat. 17 , 2965–2985. https://doi.org/10.2147/NDT.S304567 (2021). Boonstra, T. A. et al. Gait disorders and balance disturbances in Parkinson's disease: clinical update and pathophysiology. Curr. Opin. Neurol. 21 , 461–471. https://doi.org/10.1097/WCO.0b013e328305bdaf (2008). Murman, D. L. Early treatment of Parkinson's disease: opportunities for managed care. Am. J. Manag Care . 18 , S183–S188 (2012). Pagan, F. L. Improving outcomes through early diagnosis of Parkinson's disease. Am. J. Manag Care . 18 , S176–S182 (2012). Rizzo, G. et al. Accuracy of clinical diagnosis of Parkinson disease: a systematic review and meta-analysis. Neurology . 86 , 566–576. https://doi.org/10.1212/WNL.0000000000002350 (2016). Demonceau, M. et al. Contribution of a trunk accelerometer system to the characterization of gait in patients with mild-to-moderate Parkinson's disease. IEEE J. Biomed. Health Inf. 19 , 1803–1808. https://doi.org/10.1109/JBHI.2015.2469540 (2015). Zhang, X. et al. Single- and dual-task gait performance and their diagnostic value in early-stage Parkinson's disease. Front. Neurol. 13 , 974985. https://doi.org/10.3389/fneur.2022.974985 (2022). Zhu, S. et al. Gait analysis with wearables is a potential progression marker in Parkinson's disease. Brain Sci. 12 , 1213. https://doi.org/10.3390/brainsci12091213 (2022). Ferreira, M. I. A. et al. Machine learning models for Parkinson's disease detection and stage classification based on spatial-temporal gait parameters. Gait Posture . 98 , 49–55. https://doi.org/10.1016/j.gaitpost.2022.08.014 (2022). Shcherbak, A. et al. IEEE,. Dominant hand invariant Parkinson's disease detection using 1-D CNN model and STFT-based IMU data fusion. In 2023 IEEE 32nd International Symposium on Industrial Electronics (ISIE) 1–6 (2023). Yang, X. et al. PD-ResNet for classification of Parkinson's disease from gait. IEEE J. Transl Eng. Health Med. 10 , 1–11. https://doi.org/10.1109/JTEHM.2022.3180933 (2022). Mirelman, A. et al. Gait impairments in Parkinson's disease. Lancet Neurol. 18 , 697–708. https://doi.org/10.1016/S1474-4422(19)30044-4 (2019). Dodel, R. C., Berger, K. & Oertel, W. H. Health-related quality of life and healthcare utilisation in patients with Parkinson's disease: impact of motor fluctuations and dyskinesias. Pharmacoeconomics . 19 , 1013–1038. https://doi.org/10.2165/00019053-200119100-00004 (2001). Caraceni, T., Scigliano, G. & Musicco, M. The occurrence of motor fluctuations in Parkinsonian patients treated long term with levodopa: role of early treatment and disease progression. Neurology . 41 , 380–384. https://doi.org/10.1212/wnl.41.3.380 (1991). Martignon, C. et al. Guidelines on exercise testing and prescription for patients at different stages of Parkinson's disease. Aging Clin. Exp. Res. 33 , 221–246. https://doi.org/10.1007/s40520-020-01612-1 (2021). Zhao, H. et al. Severity level diagnosis of Parkinson's disease by ensemble K-nearest neighbor under imbalanced data. Expert Syst. Appl. 189 , 116113. https://doi.org/10.1016/j.eswa.2021.116113 (2022). Lee, M., Youm, C., Noh, B. & Park, H. Gait characteristics based on shoe-type inertial measurement units in healthy young adults during treadmill walking. Sensors . 20 , 2095 (2020). Almarwani, M. et al. Challenging the motor control of walking: gait variability during slower and faster pace walking conditions in younger and older adults. Arch. Gerontol. Geriatr. 66 , 54–61. https://doi.org/10.1016/j.archger.2016.05.001 (2016). Hughes, A. J., Ben-Shlomo, Y., Daniel, S. E. & Lees, A. J. UK Parkinson's Disease Society Brain Bank clinical diagnostic criteria. J. Neurol. Neurosurg. Psychiatry . 55 , e4 (1992). Folstein, M. F., Folstein, S. E. & McHugh, P. R. Mini-mental state': a practical method for grading the cognitive state of patients for the clinician. J. Psychiatr Res. 12 , 189–198. https://doi.org/10.1016/0022-3956(75)90026-6 (1975). Lee, M. et al. Validity of shoe-type inertial measurement units for Parkinson's disease patients during treadmill walking. J. Neuroeng. Rehabil . 15 , 38. https://doi.org/10.1186/s12984-018-0384-9 (2018). Nonnekes, J. et al. Compensation strategies for gait impairments in Parkinson disease: a review. JAMA Neurol. 76 , 718–725 (2019). Winter, D. A. et al. Biomechanics and motor control of human gait: normlderly and pathological (1991). Plotnik, M., Giladi, N. & Hausdorff, J. M. A new measure for quantifying the bilateral coordination of human gait: effects of aging and Parkinson's disease. Exp. Brain Res. 181 , 561–570 (2007). Bland, J. M. & Altman, D. G. Measuring agreement in method comparison studies. Stat. Methods Med. Res. 8 , 135–160 (1999). Fischer, J. E., Bachmann, L. M. & Jaeschke, R. A readers' guide to the interpretation of diagnostic test properties: clinical example of sepsis. Intensive Care Med. 29 , 1043–1051. https://doi.org/10.1007/s00134-003-1761-8 (2003). Fan, R. E., Chang, K. W., Hsieh, C. J., Wang, X. R. & Lin, C. J. LIBLINEAR: A library for large linear classification. J. Mach. Learn. Res. 9 , 1871–1874 (2008). Fix, E., Hodges, J. L. & Discriminatory analysis. Nonparametric discrimination: consistency properties. Int. Stat. Rev . 57, 238–247 (1989). Zhang, H. Exploring conditions for the optimality of naive Bayes. Int. J. Pattern Recognit. Artif. Intell. 19 , 183–198 (2005). Hastie, T., Tibshirani, R. & Friedman, J. H. The elements of statistical learning: data mining, inference, and prediction (Springer, 2009). Bishop, C. M. Pattern recognition and machine learning (Springer, 2006). Breiman, L. Random forests. Mach. Learn. 45 , 5–32 (2001). Lemaitre, G. et al. Computer-aided detection for prostate cancer detection based on multi-parametric magnetic resonance imaging. In. 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) 3138–3141 (IEEE, 2017). (2017). https://doi.org/10.1109/EMBC.2017.8037522 Zanardi, A. P. J. et al. Gait parameters of Parkinson's disease compared with healthy controls: a systematic review and meta-analysis. Sci. Rep. 11 , 752. https://doi.org/10.1038/s41598-020-80768-2 (2021). Von der Recke, F. et al. Reduced range of gait speed: a Parkinson's disease-specific symptom? J. Parkinsons Dis. 13 , 197–202. https://doi.org/10.3233/JPD-223535 (2023). Trabassi, D. et al. Machine learning approach to support the detection of Parkinson's disease in IMU-based gait analysis. Sensors . 22 , 3700. https://doi.org/10.3390/s22103700 (2022). Mak, M. K. Reduced step length, not step length variability, is central to gait hypokinesia in people with Parkinson's disease. Clin. Neurol. Neurosurg. 115 , 587–590. https://doi.org/10.1016/j.clineuro.2012.07.014 (2013). Schniepp, R. et al. Locomotion speed determines gait variability in cerebellar ataxia and vestibular failure. Mov. Disord . 27 , 125–131. https://doi.org/10.1002/mds.23978 (2012). Rennie, L. et al. The reliability of gait variability measures for individuals with Parkinson's disease and healthy older adults–The effect of gait speed. Gait Posture . 62 , 505–509. https://doi.org/10.1016/j.gaitpost.2018.04.011 (2018). Cole, M. H. et al. Imposed faster and slower walking speeds influence gait stability differently in Parkinson fallers. Arch. Phys. Med. Rehabil . 98 , 639–648 (2017). Hartmann, A., Luzi, S., Murer, K., de Bie, R. A. & de Bruin E. D. Concurrent validity of a trunk tri-axial accelerometer system for gait analysis in older adults. Gait Posture . 29 , 444–448 (2009). Monaghan, K., Delahunt, E. & Caulfield, B. Increasing the number of gait trial recordings maximises intra-rater reliability of the CODA motion analysis system. Gait Posture . 25 , 303–315. https://doi.org/10.1016/j.gaitpost.2006.04.011 (2007). Vila, M. H., Pérez, R., Mollinedo, I. & Cancela, J. M. Analysis of gait for disease stage in patients with Parkinson's disease. Int. J. Environ. Res. Public. Health . 18 , 720. https://doi.org/10.3390/ijerph18020720 (2021). Canning, C. G., Ada, L., Johnson, J. J. & McWhirter, S. Walking capacity in mild to moderate Parkinson's disease. Arch. Phys. Med. Rehabil . 87 , 371–375 (2006). Chatzaki, C. et al. Can gait features help in differentiating Parkinson's disease medication states and severity levels? A machine learning approach. Sensors . 22 , 9937 (2022). Curtze, C., Nutt, J. G., Carlson-Kuhta, P., Mancini, M. & Horak, F. B. Levodopa is a double-edged sword for balance and gait in people with Parkinson's disease. Mov. Disord . 30 , 1361–1370 (2015). Vabalas, A., Gowen, E., Poliakoff, E. & Casson, A. J. Machine learning algorithm validation with a limited sample size. PLoS One . 14 , e0224365 (2019). Albrecht, F. et al. Unraveling Parkinson's disease heterogeneity using subtypes based on multimodal data. Parkinsonism Relat. Disord . 102 , 19–29 (2022). Tables Table 1 . Physical and clinical characteristics of all participants. Characteristics Controls ( n = 75) Early PDs ( n = 61) Moderate PDs ( n = 42) p -value Sex (male/female) 33/42 36/25 28/14 0.050 Age (years) 68.91±5.54 68.61±6.24 69.45±5.53 0.766 Height (cm) 160.61±8.97 161.89±8.46 160.15±8.72 0.560 Body mass (kg) 63.94±9.94 64.09±9.28 61.65±10.31 0.397 BMI (kg/m 2 ) 24.75±3.13 24.41±2.64 23.89±2.29 0.279 K-MMSE (scores) 27.28±1.80 27.75±1.71 27.6±1.53 0.253 Hoehn and Yahr scale (stages) - 1.84±0.37 2.69±0.25 <0.001 a UPDRS total (scores) - 52.00±16.27 66.81±17.83 <0.001 b UPDRS part Ⅲ (scores) - 32.84±12.20 42.23±13.07 <0.001 b PIGD (scores) - 0.60±0.34 1.17±0.59 <0.001 a L-Dopa equivalent dose (mg/day) - 571.89±330.91 917.48±519.66 <0.001 a Treatment duration (years) - 4.93±3.43 6.33±3.30 0.020 a Symptom duration (years) - 5.36±3.39 7.12±3.76 0.012 a All data are indicated as mean ± standard deviation; PDs: Parkinson’s disease; BMI: Body mass index; K-MMSE: Korean mini-mental state examination; UPDRS: Unified Parkinson’s disease rating score; PIGD: Postural instability/gait difficulty. a Mann–Whitney U test; b Independent sample T-test; p < 0.05. Table 2 . Model parameters of seven classifiers estimated by grid search. MLs Controls vs. PDs (30 Features) Controls vs. PDs (3 Features) Controls vs. Early PDs (30 Features) Controls vs. Early PDs (2 Features) Early PDs vs. Moderate PDs (30 Features) Early PDs vs. Moderate PDs (1 Feature) LR C = 1.0 C = 1.0 C = 0.001 C = 0.1 C = 1.0 C = 0.1 KNN k = 9 k = 9 k = 8 k = 7 k = 3 k = 7 NB - - - - - - LDA n_components=1 n_components=1 n_components=1 n_components=1 n_components=1 n_components=1 QDA reg_param = 0.3 reg_param=0.001 reg_param=0.001 reg_param = 0.4 reg_param = 0.1 reg_param = 0.5 SVM C = 7.6 gamma = scale, kernel = linear C = 29.5, gamma = 0.01, kernel = rbf C = 1.9, gamma = 0.001, kernel = rbf C = 1719, gamma = 0.0001, kernel = rbf C = 0.5, gamma = 1.0, kernel = rbf C = 7.6, gamma = 0.01, kernel = rbf RF max_depth=10, n_estimators=1250 max_depth=30, n_estimators=1250 max_depth=15, n_estimators=1500 max_depth=15, n_estimators=500 max_depth=30, n_estimators=750 max_depth=10, n_estimators=500 ML: Machine learning; PDs: People with Parkinson’s disease; LR: Logistic regression, “C” is the inverse of the regularization strength; KNN: K-nearest neighbor, “ k ” is the number of neighbors; NB: Naïve Bayes; LDA: Linear discriminant analysis, “n_components” is the number of components; QDA: Quadratic discriminant analysis, “reg_param” is the regularization of the per-class covariance; SVM: Support vector machine, “C” is the regularization parameter, and “gamma” is the kernel coefficient; RF: Random forest, “n_estimators” is the number of trees in the forest, and “max_depth” is the maximum depth of the tree. Table 3 . Results of binary logistic regression for different groups analyzed. Variables Estimate SE OR 95% CI for OR p-value R N 2 Controls and PDs CV of stride length at the LWS (%) 0.599 0.256 1.820 1.101–3.007 0.020 0.528 Walking speed at the PWS (m/s) −0.864 0.392 0.421 0.195–0.909 0.027 Stride length at the HWS (m) −1.642 0.460 0.194 0.079–0.477 < 0.001 Controls and early PDs CV of stride length at the LWS (%) 0.535 0.240 1.707 1.067 – 2.732 0.026 0.404 Stride length at the HWS (m) −1.790 0.364 0.167 0.082 – 0.341 < 0.001 Early PDs and moderate PDs Walking speed at the PWS (m/s) −0.979 0.282 0.376 0.216 – 0.653 0.001 0.291 Dependent variable 1 = Controls, 2 = PDs; 1 = Controls, 2 = H&Y 1–2; 1 = H&Y 1–2, 2 = H&Y 2.5–3. This model adjusted for age, gender, height, BMI, and MMSE score; PDs: People with Parkinson’s disease; Early PDs: H&Y 1–2; Moderate PDs: H&Y 2.5–3; CV: Coefficient of variance; PWS: Preferred walking speed; LWS: Lower walking speed; HWS: Higher walking speed; SE: Standard error; OR: Odds ratio; CI: Confidence interval; R N 2 is the fit statistic for the Nagelkerke model; p < 0.05. Table 4 . Accuracies of seven classifiers from five-fold cross-validation. MLs Controls vs. PDs (30 Features) Controls vs. PDs (3 Features) Controls vs. Early PDs (30 Features) Controls vs. Early PDs (2 Features) Early PDs vs. Moderate PDs (30 Features) Early PDs vs. Moderate PDs (1 Feature) LR 67.9 ± 9.1 69.4 ± 6.7 65.3 ± 4.5 66.0 ± 4.3 64.8 ± 14.1 68.1 ± 9.3 KNN 64.6 ± 6.4 67.9 ± 6.4 62.7 ± 5.5 64.0 ± 6.0 70.5 ± 11.6 63.9 ± 11.9 NB 66.5 ± 7.8 68.9 ± 7.7 63.3 ± 11.3 67.3 ± 8.6 56.7 ± 12.9 69.0 ± 10.6 LDA 65.5 ± 8.9 68.4 ± 8.7 64.7 ± 9.3 64.7 ± 3.0 57.5 ± 8.7 68.1 ± 9.3 QDA 68.9 ± 6.5 69.9 ± 7.9 66.7 ± 5.3 66.0 ± 8.3 64.8 ± 6.9 69.8 ± 11.4 SVM 66.0 ± 8.1 66.5 ± 9.0 67.3 ± 4.9 64.7 ± 3.8 75.5 ± 5.4 69.8 ± 11.4 RF 79.1 ± 5.4 78.1 ± 9.0 71.3 ± 6.9 61.3 ± 11.2 75.4 ± 6.4 66.5 ± 10.2 Mean (%) ± standard deviation (%) is calculated from the five-fold cross-validation; the mean values presented in boldface denote the best performance (the highest test accuracy); MLs: Machine learning techniques; LR: Logistic regression; KNN: K-nearest neighbors; NB: Naїve Bayes; LDA: Linear discriminant analysis; QDA: Quadratic discriminant analysis; SVM: Support vector machine; RF: Random forest. Additional Declarations No competing interests reported. Supplementary Files SupplementaryFigure1.pdf Supplementary Figure 1. Walking test protocol. (a) Definition of walking speeds, (b) Data collection and analysis phase for data collected from the shoe-type data logger. SupplementaryFigure2.pdf Supplementary Figure 2. Heatmap representing the importance value of all variables for three classification cases: the deeper green, to indicate higher importance. LWS: Lower walking speed; PWS: Preferred walking speed; HWS: Higher walking speed; CV: Coefficient of variance. SupplementaryTableS1.docx Supplementary Table S1. Reliability of the results for lower and higher walking speeds. SupplementaryTableS2.docx Supplementary Table S2. Variable Importance Assessment with Random Forest. SupplementaryTableS3.docx Supplementary Table S3. Comparison of gait variables between age-matched healthy controls and people with PD. SupplementaryTableS4.docx Supplementary Table S4. Comparison of gait variables between age-matched healthy controls and Early stage PD. SupplementaryTableS5.docx Supplementary Table S5. Comparison of gait variables between early and moderate stages of PD. SupplementaryTableS6.docx Supplementary Table S6. Precision, recall, and F1 score of the seven classifiers for six cases. Supplementaryfile1.Rawdata.xlsx Cite Share Download PDF Status: Published Journal Publication published 02 Jan, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 20 Oct, 2024 Reviews received at journal 17 Oct, 2024 Reviews received at journal 11 Oct, 2024 Reviewers agreed at journal 10 Oct, 2024 Reviewers agreed at journal 10 Oct, 2024 Reviewers invited by journal 09 Oct, 2024 Editor assigned by journal 09 Oct, 2024 Editor invited by journal 04 Oct, 2024 Submission checks completed at journal 03 Oct, 2024 First submitted to journal 03 Oct, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5195774","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":368293819,"identity":"306fa504-a014-4720-83a9-e8b566497381","order_by":0,"name":"Juseon Hwang","email":"","orcid":"","institution":"The Graduate School of Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Juseon","middleName":"","lastName":"Hwang","suffix":""},{"id":368293820,"identity":"64552f2d-f847-434c-9352-3f79255d7d39","order_by":1,"name":"Changhong Youm","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYDACCTBpw9gA4SYQrSWNdC2HSdDCP7vH8DFvznnZ/hkJjB9+MKTlE7bkzhljY95tt41n3EhgluxhyLFsIKTFQCLHTBqoJbHhRgKDNANDhQFBW6BaziXOB9rymxQtBxI33EhgA9qSQ1iLxI20YsO525KNN5552GbZY5BGWAv/jOSND95us5Oddzz58I0fFcmEtTAwcMAUgaKGGA0MDOwPiFI2CkbBKBgFIxgAAMSZOXu0+WMMAAAAAElFTkSuQmCC","orcid":"","institution":"The Graduate School of Dong-A University","correspondingAuthor":true,"prefix":"","firstName":"Changhong","middleName":"","lastName":"Youm","suffix":""},{"id":368293821,"identity":"d6ee21c6-3f26-40e2-b898-58798c3ed2b4","order_by":2,"name":"Hwayoung Park","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hwayoung","middleName":"","lastName":"Park","suffix":""},{"id":368293823,"identity":"acd6be57-381e-4323-88ac-331dc019d695","order_by":3,"name":"Bohyun Kim","email":"","orcid":"","institution":"The Graduate School of Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Bohyun","middleName":"","lastName":"Kim","suffix":""},{"id":368293830,"identity":"dd178478-e506-4807-a8ee-9f8223838378","order_by":4,"name":"Hyejin Choi","email":"","orcid":"","institution":"The Graduate School of Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Hyejin","middleName":"","lastName":"Choi","suffix":""},{"id":368293831,"identity":"9a0084f0-b141-43e8-b6ff-e81430b8a235","order_by":5,"name":"Sang-Myung Cheon","email":"","orcid":"","institution":"Dong-A University","correspondingAuthor":false,"prefix":"","firstName":"Sang-Myung","middleName":"","lastName":"Cheon","suffix":""}],"badges":[],"createdAt":"2024-10-03 04:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5195774/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5195774/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-83975-3","type":"published","date":"2025-01-02T15:56:57+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":69060794,"identity":"0ac668fa-8be4-4374-a8fc-2d19b9ebd31f","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":67551,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlow diagram for participants.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1781.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/4d1a3ce19469ac3569fc710f.png"},{"id":69063490,"identity":"3971c22b-0894-4d2a-bebb-d934c56db749","added_by":"auto","created_at":"2024-11-15 08:09:34","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":528942,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eData process for machine learning analysis.\u003c/strong\u003eLR: Logistic regression; KNN: K­nearest neighbors; NB: Naїve Bayes; LDA: Linear discriminant analysis; QDA: Quadratic discriminant analysis; SVM: Support vector machine; RF: Random forest; PDs: people with PD; Early PDs: people with PD in the early stage; Moderate PDs: people with PD in the moderate stage\u003c/p\u003e","description":"","filename":"Figure1782.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/fc5a01433863bfabc74dc690.png"},{"id":69062664,"identity":"49469d4c-a11c-4718-9b64-9ded8593d339","added_by":"auto","created_at":"2024-11-15 08:01:34","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":291030,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBland–Altman plots for data agreement between estimated and measured walking speeds.\u003c/strong\u003e (a) and (b) lower speed in controls and people with PD, (c) and (d) higher speed in controls and people with PD; LOA: limits of agreement.\u003c/p\u003e","description":"","filename":"Figure1783.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/fdac5018244ef858c318a02d.png"},{"id":69061886,"identity":"bcd9c55c-635e-4db1-b738-cf010694b571","added_by":"auto","created_at":"2024-11-15 07:45:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":303910,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eResults of all 30 variables importance\u003c/strong\u003e. The red dotted line surrounds the variables that were ranked in the top 15; LWS: Lower walking speed; PWS: Preferred walking speed; HWS: Higher walking speed; CV: Coefficient of variance.\u003c/p\u003e","description":"","filename":"Figure1784.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/5724a7c22243049a2d448cea.png"},{"id":69060809,"identity":"053355de-a3d1-4d3a-ba65-c4237c3d2886","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":478795,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAccuracies of seven classifiers.\u003c/strong\u003e Orange line in box plot indicates mean values. LR: Logistic regression; KNN: K­nearest neighbors; NB: Naїve Bayes; LDA: Linear discriminant analysis; QDA: Quadratic discriminant analysis; SVM: Support vector machine; RF: Random forest.\u003c/p\u003e","description":"","filename":"Figure1785.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/d093030e3d9006cbe0da0267.png"},{"id":69061888,"identity":"e53c3aae-8428-4ed1-9fdf-8ce2bb0b5b87","added_by":"auto","created_at":"2024-11-15 07:45:34","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":308907,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion matrices for six cases.\u003c/strong\u003e PDs: people with PD; RF: Random forest; NB: Naїve Bayes; SVM: Support vector machine.\u003c/p\u003e","description":"","filename":"Figure1786.png","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/166cf0f97ac6ed3623819705.png"},{"id":73093134,"identity":"5c62a0cd-937a-44aa-a42b-5dcec1317d16","added_by":"auto","created_at":"2025-01-06 16:05:28","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2642302,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/bb903053-82ff-4db0-8859-f2696c65d3bd.pdf"},{"id":69060800,"identity":"d1cbd76a-dba7-48fc-872f-eeea7c3615fa","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":170516,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Figure 1. Walking test protocol.\u003c/strong\u003e (a) Definition of walking speeds, (b) Data collection and analysis phase for data collected from the shoe-type data logger.\u003c/p\u003e","description":"","filename":"SupplementaryFigure1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/b1b65df21007eb4591e96cc3.pdf"},{"id":69063491,"identity":"07bcd0f1-57e7-44be-8659-1aeb868ab4ec","added_by":"auto","created_at":"2024-11-15 08:09:34","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":369459,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Figure 2. Heatmap representing the importance value of all variables for three classification cases: the deeper green, to indicate higher importance.\u003c/strong\u003e LWS: Lower walking speed; PWS: Preferred walking speed; HWS: Higher walking speed; CV: Coefficient of variance.\u003c/p\u003e","description":"","filename":"SupplementaryFigure2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/539eec25c09cf3c1de6d9ab9.pdf"},{"id":69060796,"identity":"31084c5e-b400-4544-928f-1705eeeff06b","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":17749,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S1.\u003c/strong\u003e Reliability of the results for lower and higher walking speeds.\u003c/p\u003e","description":"","filename":"SupplementaryTableS1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/5a8dd0a45de300bde83a8642.docx"},{"id":69061879,"identity":"e5419e8a-f373-49bc-8c76-8f75fdc8a5ea","added_by":"auto","created_at":"2024-11-15 07:45:34","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":22455,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S2.\u003c/strong\u003e Variable Importance Assessment with Random Forest.\u003c/p\u003e","description":"","filename":"SupplementaryTableS2.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/797df5bed97143b6c17f04e0.docx"},{"id":69062254,"identity":"20781eaa-f380-4db2-b1c3-215bc762db58","added_by":"auto","created_at":"2024-11-15 07:53:34","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":22521,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S3.\u003c/strong\u003e Comparison of gait variables between age-matched healthy controls and people with PD.\u003c/p\u003e","description":"","filename":"SupplementaryTableS3.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/573c04d608bef9044d38bf50.docx"},{"id":69060803,"identity":"36fd5b04-4176-420c-a378-565b4e34be85","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":23247,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S4.\u003c/strong\u003e Comparison of gait variables between age-matched healthy controls and Early stage PD.\u003c/p\u003e","description":"","filename":"SupplementaryTableS4.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/0213c1b42682f91b20140500.docx"},{"id":69060805,"identity":"650dcff8-c36b-498e-b589-508c4e47f758","added_by":"auto","created_at":"2024-11-15 07:37:34","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":23368,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S5.\u003c/strong\u003e Comparison of gait variables between early and moderate stages of PD.\u003c/p\u003e","description":"","filename":"SupplementaryTableS5.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/afc8ad6cc32a6b505e1f818f.docx"},{"id":69061884,"identity":"ef8d8c57-31d4-408c-bfdf-84cadda92bb7","added_by":"auto","created_at":"2024-11-15 07:45:34","extension":"docx","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":22767,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table S6.\u003c/strong\u003e Precision, recall, and F1 score of the seven classifiers for six cases.\u003c/p\u003e","description":"","filename":"SupplementaryTableS6.docx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/4b2e1aab86c8f101a63628d5.docx"},{"id":69062257,"identity":"4c3c6caa-4380-4d8c-a46b-e8b502c6cafc","added_by":"auto","created_at":"2024-11-15 07:53:34","extension":"xlsx","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":84846,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile1.Rawdata.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-5195774/v1/93168a21f25c089174436174.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning for Early Detection and Severity Classification in People with Parkinson's Disease","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eParkinson\u0026rsquo;s disease (PD) is a progressive neurodegenerative disorder characterized by the loss of dopaminergic neurons in the substantia nigra of the midbrain, which disrupts the function of the basal ganglia circuitry and impairs motor control [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. One of the key motor symptoms of PD is the reduced ability to perform automated movements, such as walking [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Consequently, individuals with PD often experience significant locomotor difficulties, leading to a loss of independence and reduced quality of life (QoL) [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Given these challenges, early management of PD is crucial. Timely intervention may help delay disease progression, alleviate both motor and non-motor symptoms, and enhance the QoL of people with PD, potentially resulting in substantial cost savings for healthcare systems [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, one of the primary obstacles to early intervention is that PD is often diagnosed at an advanced stage [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Additionally, distinguishing early-stage PD from normal age-related gait impairments is challenging, as both conditions may present with similar symptoms, such as decreased mobility and impaired balance, due to age-related degeneration in musculoskeletal and neurological functions [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Moreover, the early symptoms of PD are often subtle, leading to delayed diagnosis and treatment until the disease has progressed and symptoms become more pronounced and difficult to manage [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. As a result, there is growing interest in using gait assessments as a potential diagnostic tool for detecting PD in its early stages. While gait assessments show promise for the early diagnosis of PD [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], further research is needed to refine these methods and improve their accuracy in distinguishing PD from other age-related conditions. Such advancements could enable earlier interventions, improve patient outcomes, and slow disease progression.\u003c/p\u003e \u003cp\u003eRecently, there has been limited research on the early detection of PD through gait assessment using machine learning (ML) and deep learning algorithms based on artificial intelligence [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. For instance, Ferreira et al. conducted a gait analysis to classify individuals with early-stage PD (Hoehn and Yahr [H\u0026amp;Y] stages 1 and 2) and healthy controls. The study demonstrated that the support vector machine (SVM) algorithm achieved an accuracy of 76.9% when utilizing all 22 gait features, whereas the Na\u0026iuml;ve Bayes (NB) classifier reached a higher accuracy of 84.6% using four selected features [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, the study had a small sample size and relied on averaged data collected from noncontinuous gait analysis. This method does not fully replicate natural walking performance, which limits the reproducibility and generalizability of the findings.\u003c/p\u003e \u003cp\u003eAs PD progresses, typical pathological symptoms such as bradykinesia, rigidity, and impaired motor automaticity become increasingly pronounced. This often necessitates higher dosages of antiparkinson medications [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, escalating dosages can lead to motor complications, including motor response fluctuations and dyskinesias [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. The decision to initiate levodopa treatment is primarily influenced by the severity and rate of progression of the disease [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Accurate classification of PD stage is crucial for optimizing therapeutic interventions, as different stages exhibit distinct motor and non-motor characteristics. Such classifications enable clinicians to tailor exercise interventions and medical treatments to the specific needs of patients at various stages of the disease [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Subtyping PD according to disease stage could facilitate the development of personalized treatment protocols, potentially reducing dependence on medications associated with significant side effects, such as dyskinesias and motor fluctuations [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn a related study, Zhao et al. assessed gait performance using only self-selected walking speed to detect the severity of PD (H\u0026amp;Y stages 2, 2.5, and 3) [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The results revealed that the ensemble K-nearest neighbor algorithm achieved an accuracy of 76.03%. However, the study faced similar limitations, and since there remains no consensus on using gait variables to detect PD severity, further research is necessary to explore gait variables that can classify moderate stages of PD. Additionally, Additionally, Lee et al. recommended assessing gait ability across various walking speeds commonly encountered in daily life to detect the severity of motor symptoms in individuals with PD [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. These more challenging walking tasks may provide a more accurate assessment of gait variability, as they demand greater executive function and attention to maintain dynamic stability [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Therefore, it is essential to consider challenging walking tasks under different speed conditions to identify markers of disease progression in individuals with PD. This study aimed to identify the accuracy of ML algorithms in classifying gait variables based on continuous gait tasks using three walking speeds, specifically to classify: i) people with PD versus age-matched healthy controls, ii) people with early PD (H\u0026amp;Y stage 1 to 2) versus age-matched healthy controls, and iii) people with moderate PD (H\u0026amp;Y stage 2.5 to 3) versus people with early PD (H\u0026amp;Y stage 1 to 2).\u003c/p\u003e"},{"header":"METHODS","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eParticipants\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 103 individuals with PD and 75 age-matched healthy controls participated in this\u0026nbsp;study. The inclusion criteria for participants with PD were: a) a\u0026nbsp;clinician-confirmed\u0026nbsp;diagnosis of idiopathic PD according to the UK Parkinson\u0026rsquo;s Disease Society Brain Bank Clinical Diagnostic Criteria [21], b) age over 50 years, c) a modified H\u0026amp;Y scale score of 1\u0026ndash;3, indicating independent walking and mobility without the need for an assistive device, and d) a score of \u0026ge; 24 on the Korean Mini-Mental State Examination [22]. The control group consisted of age-matched healthy adults over 50 years old with no history of musculoskeletal, cardiovascular, vestibular, or neurological disorders that could have affected gait in the past six\u0026nbsp;months (Figure 1). The physical and clinical characteristics of all participants are presented in Table 1. Group 1 consisted of 75 age-matched healthy controls. Participants with PD were further categorized based on their H\u0026amp;Y scale: Group 2 included 61 those in stages 1 to 2, and Group 3 included 42 those in stages 2.5 to 3. This study was approved by the Institutional Review Board of Dong-A University Hospital (IRB number: DAUHIRB-22-089).\u0026nbsp;All\u0026nbsp;participants were informed of the experimental procedures and provided written informed consent, as approved by the committee. The study is registered in the Clinical Research Information Service\u0026nbsp;of\u0026nbsp;the Republic of Korea (KCT0009353).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTest Procedures and Data Analysis\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGait analysis in this study was conducted using shoe-type data loggers (Smart Balance\u0026reg; SB-1, JEIOS, Korea) equipped with an inertial measurement unit (IMU) embedded in the shoe outsole [23]. The IMU sensor captured triaxial acceleration (up to \u0026plusmn;6 g) and angular velocity (up to \u0026plusmn;500\u0026deg;/s) along three orthogonal axes. The measured data were transmitted via Bluetooth to a gait analysis system (DynaStab\u0026trade;, JEIOS, Korea). Participants with PD performed the gait test while in the \u0026quot;ON\u0026quot; state of antiparkinsonian medication for at least 2 h, reflecting their typical gait performance during daily activities.\u003c/p\u003e\n\u003cp\u003eAll participants were first assessed by a clinician for clinical characteristics using the Korean Mini-Mental State Examination,\u0026nbsp;the\u0026nbsp;H\u0026amp;Y scale, and\u0026nbsp;the\u0026nbsp;Unified PD Rating Scale. Physical characteristics, including height, weight, shoulder width, hip and waist circumference, thigh and calf circumference, and leg length, were also measured. Participants\u0026nbsp;began with a 5-minute warm-up\u0026nbsp;involving\u0026nbsp;stretching\u0026nbsp;exercises\u0026nbsp;before\u0026nbsp;starting\u0026nbsp;the experiment. Following the warm-up, participants performed the gait test while wearing shoes embedded\u0026nbsp;with\u0026nbsp;IMU sensors on a 24-meter straight walkway. Each participant was instructed to walk at a comfortable pace, with a metronome used to measure their self-selected walking speed in steps per minute (beats/min). Lower walking speed (LWS) and higher walking speed (HWS) were defined as 80% and 120% of the preferred walking speed (PWS), respectively [19].\u0026nbsp;Prior to\u0026nbsp;the test, participants completed three practice trials. After a 5-minute break to wash out the effects of metronome-based pacing, participants\u0026nbsp;performed the gait test at both HWS and LWS. All walking speeds were\u0026nbsp;measured\u0026nbsp;using\u0026nbsp;the\u0026nbsp;metronome (beats per minute),\u0026nbsp;but the\u0026nbsp;actual gait test was conducted without\u0026nbsp;the metronome, as external auditory cues could affect the performance of\u0026nbsp;individuals with PD [24] (Supplementary Figure 1).\u003c/p\u003e\n\u003cp\u003eAll gait test data were collected at 100 Hz and filtered using a second-order Butterworth low-pass filter with a cut-off frequency of 10 Hz. Spatiotemporal gait variables were analyzed over a 20-meter section of the walkway, excluding the 2-meter acceleration and deceleration phases at either end. The spatiotemporal gait variables were calculated as follows: gait speed (m/s), stride length (m), single-support phase (%), double-support phase (%), stance phase (%), the coefficient of variation (CV) for all spatiotemporal variables [25], and gait asymmetry [26].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eStatistical\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e\u0026nbsp;Analysis\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data were checked for normality using the Shapiro\u0026ndash;Wilk test, and statistical analyses were performed using both parametric and nonparametric tests. The mean and standard deviation (SD) were calculated using one-way analysis of variance (ANOVA) for the physical characteristics of all participants, while independent t-tests and Mann\u0026ndash;Whitney U tests were used to assess the clinical characteristics of individuals with PD. An intraclass correlation coefficient (ICC) analysis (2,1) was conducted to validate the agreement between the estimated and actual measured speeds under high- and low-speed conditions. The limits of agreement were calculated using a Bland\u0026ndash;Altman plot to examine the error between the estimated and actual measured speeds [27].\u003c/p\u003e\n\u003cp\u003eTo reduce the dimensionality of the input data, this study used the random forest (RF) method to assess feature importance. Specifically, the hyperparameters for the RF algorithms were tuned by experimenting with the number of trees set to 100, 200, and 300, and the maximum number of features set to 0.25, 0.5, and 0.75, while the random state was fixed at 42. This approach helps exclude the influence of irrelevant variables and facilitates data post-processing. We evaluated the importance of variables in three cases: (i) between age-matched healthy controls and individuals with PD, (ii) between age-matched healthy controls and individuals with early-stage PD, and (iii) between individuals with early-stage PD and those with moderate-stage PD.\u003c/p\u003e\n\u003cp\u003eWe ranked the variables in order of importance for each case and selected the top 15 spatiotemporal gait variables. Finally, we used multivariate logistic forward stepwise regression analysis to identify the optimal gait variables for classifying the three groups from these top 15 variables. Covariates included participants\u0026apos; age, gender, height, and body mass index. The cut-off values of the selected gait variables for group classification were examined using receiver operating characteristic (ROC) curves. The area under the ROC curve (AUC) was calculated to assess the ability of the selected variables to classify the groups [28]. Statistical significance was set at 0.05.\u003c/p\u003e\n\u003cp\u003eAdditionally, a ML approach was applied to explore the potential of spatiotemporal gait features for classifying individuals with PD versus age-matched controls, as well as for distinguishing between different PD severity levels. Several ML algorithms were employed to evaluate the suitability of these gait variables for classification across different groups, with each algorithm providing unique insights into the data, thus necessitating an assessment of variation in classification accuracy. To address the classification tasks, seven ML algorithms were\u0026nbsp;applied (Figure 2): logistic regression (LR) [29], K-nearest neighbor (KNN) [30], NB [31], linear discriminant analysis (LDA) [32], quadratic discriminant analysis (QDA) [32], SVM [33], and random forest (RF) [34]. The model hyperparameters for each classifier were tuned using a grid search (Table 2).\u003c/p\u003e\n\u003cp\u003eFive-fold cross-validation was used to evaluate the accuracy, precision, recall, and F1 scores for the classification performance of the seven algorithms. In this method, the dataset is divided into five parts, with four parts used as training data and one part used as test data in each iteration. The process is repeated five times, and the average of the results from the test sets is used as the final accuracy. Given the imbalanced dataset, which included 75 individuals in Group 1, 61 in Group 2, and 42 in Group 3, a random oversampling method was employed to address this imbalance [35]. Statistical analyses were performed using SPSS 21.0 (IBM Corp., Armonk, NY, USA) and Python (Python 3.10, Python Software Foundation).\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eReliability of Estimated and Measured Speeds\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary Table S1 presents the reliability of the estimated and measured walking speeds at the LWS and HWS for both the control group and individuals with PD. The agreement between the estimated and measured walking speeds ranged from 92.0% to 95.1% at LWS and from 94.7% to 96.1% at HWS for both\u0026nbsp;groups (Figure 3).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFeature Selection for Group Classification\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe feature importance values for all 30 gait variables in each of the three classification cases are listed in Supplementary Table S2 (see Supplementary Tables S3, S4, and S5 for a comparison of gait variables across classification cases). The ranking of the importance of all variables, highlighting those ranked in the top 15 for each case (Figure 4,\u0026nbsp;Supplementary Figure 2). Using these selected variables, the multivariate logistic forward stepwise regression analyses revealed significant differences between the control group and individuals with PD.\u0026nbsp;Specifically, the CV for stride length at LWS (cut-off value: 2.64%; AUC: 0.409; sensitivity: 0.453; specificity: 0.456; odds ratio [OR]: 1.820), walking speed at PWS (cut-off value: 1.27 m/s; AUC: 0.781; sensitivity: 0.747; specificity: 0.738; OR: 0.421), and stride length at HWS (cut-off value: 1.43 m; AUC: 0.757; sensitivity: 0.680; specificity: 0.680; OR: 0.460) showed significant differences. In the classification\u0026nbsp;between\u0026nbsp;controls and individuals with early-stage PD, significant differences were observed in the CV of stride length at LWS (cut-off value: 2.62%; AUC: 0.405; sensitivity: 0.453; specificity: 0.459; OR: 1.707) and stride length at HWS (cut-off value: 1.48 m; AUC: 0.690; sensitivity: 0.613; specificity: 0.656; OR: 0.167). Additionally, in the classification\u0026nbsp;between\u0026nbsp;individuals with early-stage PD and\u0026nbsp;those with\u0026nbsp;moderate-stage PD, walking speed at PWS (cut-off value: 1.18 m/s; AUC: 0.739; sensitivity: 0.590; specificity: 0.738; OR: 0.376)\u0026nbsp;showed significant differences\u0026nbsp;(Table 3).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eML Approach\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study addressed three classification tasks using two distinct feature sets: one consisting of all spatiotemporal features, and the other comprising features selected through binary stepwise regression (Table 4, Figure 5). In the classification of controls and individuals with PD, the RF algorithm demonstrated the highest accuracy with all features (79.1% \u0026plusmn; 5.4% SD). After feature reduction through stepwise regression, RF maintained the highest accuracy (78.1% \u0026plusmn; 9.0% SD). For the early-stage classification of controls and individuals with PD, RF again achieved the highest accuracy (71.3% \u0026plusmn; 6.9% SD) with all features, while NB achieved the highest accuracy (67.3% \u0026plusmn; 8.6% SD) using two selected features. In classifying individuals with early-stage PD and moderate-stage PD, the SVM demonstrated the highest accuracy (75.5% \u0026plusmn; 5.4% SD) with all features, while both QDA and SVM achieved the highest accuracy (69.8% \u0026plusmn; 11.4% SD) with one selected feature. Additionally, classification performance metrics such as recall, precision, and F1 score were evaluated using a confusion matrix (Supplementary Table S6, Figure 6).\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eClassification of People with PD and Age-Matched Healthy Controls\u003c/h2\u003e \u003cp\u003eThis study classified age-matched healthy controls and individuals with PD using gait variables under LWS, PWS, and HWS conditions. We demonstrated that individuals with PD exhibited a 57.9% reduction in walking speed at PWS and a 54.0% shortening of stride length at HWS compared to controls, which is consistent with previous reports [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. Recently, Trabassi et al. evaluated the classification performance of ML models applied to individuals with PD and healthy controls, with the SVM achieving an accuracy of 81% for all 22 features and 86% for seven selected features, including the CV of step length, stride length, and stance phase [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. In comparison, the present study achieved an accuracy of 79.1% for all 30 features and 78.1% for the three selected features (CV of stride length at LWS, walking speed at PWS, and stride length at HWS) using the RF algorithm, demonstrating similar performance to previous findings.\u003c/p\u003e \u003cp\u003eIndividuals with PD exhibit disrupted walking speeds, primarily due to difficulties in generating adequately sized steps and delays in movement timing, which are attributed to impaired cue production in the basal ganglia. This disruption may also involve the under-scaling of movement through the cortical-thalamic-basal ganglia circuitry [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. However, none of the previous studies considered variable walking speeds or highlighted the importance of gait variability in their analyses. The result could be explained by the fact that walking at a preferred speed represents a gait pattern that minimizes variability and requires the least energy expenditure, regardless of cerebellar motor dysfunction or vestibular feedback control disorders [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this study, we identified the variability of stride length at HWS as an important variable for distinguishing individuals with PD from controls. Increased variability in individuals with PD is thought to stem from a reduced ability to control gait timing, resulting from dysfunction in various components of the basal ganglia and other regions of the locomotor network. This dysfunction leads to a series of disjointed strides, which ultimately decrease walking speed [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Furthermore, both LWS and HWS are associated with prolonged single-limb stance phases and lateral displacement of the center of mass, further contributing to increased gait variability in individuals with PD [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Collectively, individuals with PD may experience difficulty walking under both slow and fast conditions, which can sensitively detect a decline in walking ability. Therefore, we suggest that evaluating gait under slower, faster, and preferred walking speeds may provide a valuable approach for overcoming these limitations [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClassification of Controls and People with Early-Stage PD\u003c/h2\u003e \u003cp\u003eAs a primary finding, the classification between individuals with early-stage PD and age-matched healthy controls, using multivariate logistic forward stepwise regression analysis, identified two significant gait features: the CV of stride length at LWS (OR: 1.707; AUC: 0.405) and shortened stride length at HWS (OR: 0.167; AUC: 0.690), with an explanatory power of 40.4%. Additionally, the RF algorithm achieved an accuracy of 71.3% using all features, while NB reached an accuracy of 67.3% with the two selected features.\u003c/p\u003e \u003cp\u003eIn this study, individuals with early-stage PD demonstrated a 1.71-fold increase in the CV of stride length at LWS and an 83.3% reduction in stride length at HWS compared to age-matched healthy controls, consistent with previous studies [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Motor symptoms in early-stage PD often manifest as unilateral parkinsonism, primarily affecting one side of the body, due to asymmetrical degeneration of the basal ganglia [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Impaired basal ganglia function can result in reduced stride length during all gait phases due to bradykinesia and decreased walking velocity in individuals with early-stage PD [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eA recent study by Ferreira et al. performed gait analysis on an 8.5 m walkway at self-selected walking speeds to classify individuals with early-stage PD (H\u0026amp;Y stages 1 and 2) [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The study found that SVM with a radial basis function kernel achieved an accuracy of 76.9% using all 22 features, while NB achieved an accuracy of 84.6% using four selected features, including walking speed, step length, step width, and the CV of step width. They suggested that the pace and variability domains of walking are the most discriminative features in classifying early-stage PD, which aligns with the findings of this study. In our study, the CV of step length at HWS and step length at LWS achieved a lower accuracy (67.3% using the NB algorithm) compared to the previous study, which used non-continuous gait analysis data. Non-continuous steps may better mimic natural walking performance but can limit reproducibility and generalizability [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Some researchers suggest that gait assessments on a 20 m walkway, with at least 30 steps, can improve the reliability and generalizability of the results [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Therefore, conducting gait assessments with at least 30 consecutive steps under various speed conditions may provide a more objective and sensitive method for classifying early-stage PD from age-matched healthy controls.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eClassification of Early and Moderate Stages of PD\u003c/h2\u003e \u003cp\u003eIn the classification between individuals with early-stage PD and those with moderate-stage PD, the selected gait feature was walking speed at PWS (OR: 0.376; AUC: 0.739), with an explanatory power of 29.1%. Additionally, the SVM algorithm achieved an accuracy of 75.5% using all features, while both SVM and QDA achieved an accuracy of 69.8% using the single selected feature.\u003c/p\u003e \u003cp\u003eAs PD progresses, motor symptoms such as bradykinesia, rigidity, and bilateral involvement become more pronounced, reducing gait asymmetry compared to early-stage PD [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In the present study, walking speed at PWS was 1.707 times lower in individuals with moderate-stage PD compared to those with early-stage PD. This finding highlights the significant discriminative power of walking speed in classifying individuals with PD between early and moderate stages. Previous studies have also demonstrated that walking speed deteriorates significantly as PD progresses, consistent with the findings of this study [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. This emphasizes the importance of slower movement as a key independent factor in walking ability during the early-to-moderate stages of PD [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRecent classification studies using ML models have not specifically investigated the accuracy of classifying individuals with PD across H\u0026amp;Y stages 1 to 3 based on spatiotemporal gait features from straight walkways. In this study, we achieved an accuracy of 69.8% for walking speed at PWS, a key feature in moderate-stage PD. Although this accuracy is slightly lower than that reported in previous studies [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e], the use of sufficient consecutive steps allowed us to achieve reliable accuracy in classifying PD in the early-to-moderate stages [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Thus, gait assessments involving at least 30 consecutive steps under various speed conditions may provide a more objective and sensitive method for classifying individuals with PD across these stages.\u003c/p\u003e \u003cp\u003eThe present study offers valuable clinical insights. First, a marker based on three walking speeds with at least 30 consecutive steps was developed for H\u0026amp;Y stage classification. This marker may aid in distinguishing early-stage PD from normal aging and in classifying the severity of motor symptoms in PD. Furthermore, it has the potential to significantly impact clinical practice by facilitating personalized therapeutic and exercise interventions tailored to individual patient characteristics. Additionally, using selected gait features across three walking speeds could help physicians save time in assessing motor symptoms and provide a more objective evaluation of gait ability in clinical settings.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, we analyzed gait ability only during dopaminergic medication intake to reflect the daily activities of individuals with PD [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. However, assessing gait performance in both the ON and OFF medication states would provide a more comprehensive evaluation, as both are crucial for understanding motor performance in PD. Second, we examined a limited dataset of individuals with early- and moderate-stage PD, which may have impacted the performance of the ML models [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. Third, while the ML algorithms achieved promising results in classifying age-matched healthy controls and individuals with PD, the performance was slightly lower when classifying between early-stage and moderate-stage PD. To address these limitations, future studies should include multimodal datasets incorporating clinical, cognitive, motor, and neuroimaging variables, which may enhance the reliability and accuracy of characterizing PD progression [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study analyzed the accuracy of ML algorithms in classifying gait variables based on a continuous gait task performed at three different walking speeds, aimed at early detection and severity classification of PD. The key gait markers for distinguishing individuals with PD from healthy controls were walking speed at PWS, stride length at HWS, and the CV of stride length at LWS. These features achieved a classification accuracy of 78.1% using the RF algorithm. For early detection of PD, stride length at HWS and the CV of stride length at LWS demonstrated strong potential, achieving an accuracy of 67.3% with the NB algorithm. Furthermore, walking speed at PWS was the most effective feature for assessing disease severity, achieving an accuracy of 69.8% using SVM and QDA.\u003c/p\u003e \u003cp\u003eThese findings suggest that assessing gait over at least 30 consecutive steps at three different walking speeds could provide an objective and reliable method for evaluating motor symptoms in PD. Such assessments could serve as a valuable tool for clinicians and rehabilitation professionals, offering insights that support the development of personalized treatment and rehabilitation strategies tailored to the progression of the disease.\u003c/p\u003e "},{"header":"Declarations","content":" \u003ch2\u003eETHICS DECLARATIONS\u003c/h2\u003e \u003cp\u003e All study procedures involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The study protocol and supplementary information files were approved by the Institutional Review Board of Dong-A University Hospital (IRB number: DAUHIRB-22-089) (see ethics approval letter in the supplementary file). All patients provided written informed consent before data collection. The study is registered with the Clinical Research Information Service in the Republic of Korea (KCT0009353).\u003c/p\u003e \u003ch2\u003eCOMPETING INTERESTS\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no conflicts of interest.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFUNDING\u003c/h2\u003e \u003cp\u003eThis work was supported by a grant [Grant No. 2022R1A2C100933711; Changhong Youm] from the National Research Foundation of Korea (NRF), funded by the Korean government (MSIT). This research was also supported [Grant No. 2022R1A6A3A0108756411; Hwayoung Park] by the Basic Science Research Program through the NRF, funded by the Ministry of Education. The funders had no role in the study design, collection, analysis and interpretation of the data, and in writing the manuscript.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJH, CY, HP, BK, HC, and SC conceived and designed the study. JH, HP, BK, HC, and SC recruited the participants. JH, CY, HP, BK, HC, and SC performed the data acquisition. JH, CY, HP, BK, HC, and SC analyzed and interpreted the data. JH, CY, HP, BK, HC, and SC drafted the article. All authors read and approved the final version of the manuscript submitted.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors thank all the participants of this study. This work was supported by a Dong-A University Research Fund.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets supporting this study\u0026rsquo;s findings are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlexander, G. E., DeLong, M. R. \u0026amp; Strick, P. L. Parallel organization of functionally segregated circuits linking basal ganglia and cortex. \u003cem\u003eAnnu. Rev. Neurosci.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 357\u0026ndash;381. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1146/annurev.ne.09.030186.002041\u003c/span\u003e\u003cspan address=\"10.1146/annurev.ne.09.030186.002041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1986).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTakakusaki, K., Tomita, N. \u0026amp; Yano, M. Substrates for normal gait and pathophysiology of gait disturbances with respect to the basal ganglia dysfunction. \u003cem\u003eJ. Neurol.\u003c/em\u003e \u003cb\u003e255\u003c/b\u003e, 19\u0026ndash;29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00415-008-4004-7\u003c/span\u003e\u003cspan address=\"10.1007/s00415-008-4004-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith, M. D., Brazier, D. E. \u0026amp; Henderson, E. J. Current perspectives on the assessment and management of gait disorders in Parkinson's disease. \u003cem\u003eNeuropsychiatr Dis. Treat.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 2965\u0026ndash;2985. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2147/NDT.S304567\u003c/span\u003e\u003cspan address=\"10.2147/NDT.S304567\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoonstra, T. A. et al. Gait disorders and balance disturbances in Parkinson's disease: clinical update and pathophysiology. \u003cem\u003eCurr. Opin. Neurol.\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e, 461\u0026ndash;471. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/WCO.0b013e328305bdaf\u003c/span\u003e\u003cspan address=\"10.1097/WCO.0b013e328305bdaf\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMurman, D. L. Early treatment of Parkinson's disease: opportunities for managed care. \u003cem\u003eAm. J. Manag Care\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, S183\u0026ndash;S188 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePagan, F. L. Improving outcomes through early diagnosis of Parkinson's disease. \u003cem\u003eAm. J. Manag Care\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, S176\u0026ndash;S182 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRizzo, G. et al. Accuracy of clinical diagnosis of Parkinson disease: a systematic review and meta-analysis. \u003cem\u003eNeurology\u003c/em\u003e. \u003cb\u003e86\u003c/b\u003e, 566\u0026ndash;576. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1212/WNL.0000000000002350\u003c/span\u003e\u003cspan address=\"10.1212/WNL.0000000000002350\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDemonceau, M. et al. Contribution of a trunk accelerometer system to the characterization of gait in patients with mild-to-moderate Parkinson's disease. \u003cem\u003eIEEE J. Biomed. Health Inf.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 1803\u0026ndash;1808. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JBHI.2015.2469540\u003c/span\u003e\u003cspan address=\"10.1109/JBHI.2015.2469540\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, X. et al. Single- and dual-task gait performance and their diagnostic value in early-stage Parkinson's disease. \u003cem\u003eFront. Neurol.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 974985. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fneur.2022.974985\u003c/span\u003e\u003cspan address=\"10.3389/fneur.2022.974985\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu, S. et al. Gait analysis with wearables is a potential progression marker in Parkinson's disease. \u003cem\u003eBrain Sci.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1213. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/brainsci12091213\u003c/span\u003e\u003cspan address=\"10.3390/brainsci12091213\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerreira, M. I. A. et al. Machine learning models for Parkinson's disease detection and stage classification based on spatial-temporal gait parameters. \u003cem\u003eGait Posture\u003c/em\u003e. \u003cb\u003e98\u003c/b\u003e, 49\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gaitpost.2022.08.014\u003c/span\u003e\u003cspan address=\"10.1016/j.gaitpost.2022.08.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShcherbak, A. et al. IEEE,. Dominant hand invariant Parkinson's disease detection using 1-D CNN model and STFT-based IMU data fusion. In \u003cem\u003e2023 IEEE 32nd International Symposium on Industrial Electronics (ISIE)\u003c/em\u003e 1\u0026ndash;6 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, X. et al. PD-ResNet for classification of Parkinson's disease from gait. \u003cem\u003eIEEE J. Transl Eng. Health Med.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 1\u0026ndash;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/JTEHM.2022.3180933\u003c/span\u003e\u003cspan address=\"10.1109/JTEHM.2022.3180933\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMirelman, A. et al. Gait impairments in Parkinson's disease. \u003cem\u003eLancet Neurol.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e, 697\u0026ndash;708. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1474-4422(19)30044-4\u003c/span\u003e\u003cspan address=\"10.1016/S1474-4422(19)30044-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDodel, R. C., Berger, K. \u0026amp; Oertel, W. H. Health-related quality of life and healthcare utilisation in patients with Parkinson's disease: impact of motor fluctuations and dyskinesias. \u003cem\u003ePharmacoeconomics\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 1013\u0026ndash;1038. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2165/00019053-200119100-00004\u003c/span\u003e\u003cspan address=\"10.2165/00019053-200119100-00004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaraceni, T., Scigliano, G. \u0026amp; Musicco, M. The occurrence of motor fluctuations in Parkinsonian patients treated long term with levodopa: role of early treatment and disease progression. \u003cem\u003eNeurology\u003c/em\u003e. \u003cb\u003e41\u003c/b\u003e, 380\u0026ndash;384. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1212/wnl.41.3.380\u003c/span\u003e\u003cspan address=\"10.1212/wnl.41.3.380\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1991).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartignon, C. et al. Guidelines on exercise testing and prescription for patients at different stages of Parkinson's disease. \u003cem\u003eAging Clin. Exp. Res.\u003c/em\u003e \u003cb\u003e33\u003c/b\u003e, 221\u0026ndash;246. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40520-020-01612-1\u003c/span\u003e\u003cspan address=\"10.1007/s40520-020-01612-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, H. et al. Severity level diagnosis of Parkinson's disease by ensemble K-nearest neighbor under imbalanced data. \u003cem\u003eExpert Syst. Appl.\u003c/em\u003e \u003cb\u003e189\u003c/b\u003e, 116113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.eswa.2021.116113\u003c/span\u003e\u003cspan address=\"10.1016/j.eswa.2021.116113\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, M., Youm, C., Noh, B. \u0026amp; Park, H. Gait characteristics based on shoe-type inertial measurement units in healthy young adults during treadmill walking. \u003cem\u003eSensors\u003c/em\u003e. \u003cb\u003e20\u003c/b\u003e, 2095 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlmarwani, M. et al. Challenging the motor control of walking: gait variability during slower and faster pace walking conditions in younger and older adults. \u003cem\u003eArch. Gerontol. Geriatr.\u003c/em\u003e \u003cb\u003e66\u003c/b\u003e, 54\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.archger.2016.05.001\u003c/span\u003e\u003cspan address=\"10.1016/j.archger.2016.05.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHughes, A. J., Ben-Shlomo, Y., Daniel, S. E. \u0026amp; Lees, A. J. UK Parkinson's Disease Society Brain Bank clinical diagnostic criteria. \u003cem\u003eJ. Neurol. Neurosurg. Psychiatry\u003c/em\u003e. \u003cb\u003e55\u003c/b\u003e, e4 (1992).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFolstein, M. F., Folstein, S. E. \u0026amp; McHugh, P. R. Mini-mental state': a practical method for grading the cognitive state of patients for the clinician. \u003cem\u003eJ. Psychiatr Res.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 189\u0026ndash;198. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0022-3956(75)90026-6\u003c/span\u003e\u003cspan address=\"10.1016/0022-3956(75)90026-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1975).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, M. et al. Validity of shoe-type inertial measurement units for Parkinson's disease patients during treadmill walking. \u003cem\u003eJ. Neuroeng. Rehabil\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e, 38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12984-018-0384-9\u003c/span\u003e\u003cspan address=\"10.1186/s12984-018-0384-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNonnekes, J. et al. Compensation strategies for gait impairments in Parkinson disease: a review. \u003cem\u003eJAMA Neurol.\u003c/em\u003e \u003cb\u003e76\u003c/b\u003e, 718\u0026ndash;725 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWinter, D. A. et al. \u003cem\u003eBiomechanics and motor control of human gait: normlderly and pathological\u003c/em\u003e (1991).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePlotnik, M., Giladi, N. \u0026amp; Hausdorff, J. M. A new measure for quantifying the bilateral coordination of human gait: effects of aging and Parkinson's disease. \u003cem\u003eExp. Brain Res.\u003c/em\u003e \u003cb\u003e181\u003c/b\u003e, 561\u0026ndash;570 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBland, J. M. \u0026amp; Altman, D. G. Measuring agreement in method comparison studies. \u003cem\u003eStat. Methods Med. Res.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 135\u0026ndash;160 (1999).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFischer, J. E., Bachmann, L. M. \u0026amp; Jaeschke, R. A readers' guide to the interpretation of diagnostic test properties: clinical example of sepsis. \u003cem\u003eIntensive Care Med.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e, 1043\u0026ndash;1051. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00134-003-1761-8\u003c/span\u003e\u003cspan address=\"10.1007/s00134-003-1761-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan, R. E., Chang, K. W., Hsieh, C. J., Wang, X. R. \u0026amp; Lin, C. J. LIBLINEAR: A library for large linear classification. \u003cem\u003eJ. Mach. Learn. Res.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 1871\u0026ndash;1874 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFix, E., Hodges, J. L. \u0026amp; Discriminatory analysis. Nonparametric discrimination: consistency properties. \u003cem\u003eInt. Stat. Rev\u003c/em\u003e. 57, 238\u0026ndash;247 (1989).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, H. Exploring conditions for the optimality of naive Bayes. \u003cem\u003eInt. J. Pattern Recognit. Artif. Intell.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 183\u0026ndash;198 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHastie, T., Tibshirani, R. \u0026amp; Friedman, J. H. \u003cem\u003eThe elements of statistical learning: data mining, inference, and prediction\u003c/em\u003e (Springer, 2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBishop, C. M. \u003cem\u003ePattern recognition and machine learning\u003c/em\u003e (Springer, 2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman, L. Random forests. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 5\u0026ndash;32 (2001).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLemaitre, G. et al. Computer-aided detection for prostate cancer detection based on multi-parametric magnetic resonance imaging. In. \u003cem\u003e39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)\u003c/em\u003e 3138\u0026ndash;3141 (IEEE, 2017). (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/EMBC.2017.8037522\u003c/span\u003e\u003cspan address=\"10.1109/EMBC.2017.8037522\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZanardi, A. P. J. et al. Gait parameters of Parkinson's disease compared with healthy controls: a systematic review and meta-analysis. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 752. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-020-80768-2\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-80768-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVon der Recke, F. et al. Reduced range of gait speed: a Parkinson's disease-specific symptom? \u003cem\u003eJ. Parkinsons Dis.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 197\u0026ndash;202. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/JPD-223535\u003c/span\u003e\u003cspan address=\"10.3233/JPD-223535\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrabassi, D. et al. Machine learning approach to support the detection of Parkinson's disease in IMU-based gait analysis. \u003cem\u003eSensors\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e, 3700. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/s22103700\u003c/span\u003e\u003cspan address=\"10.3390/s22103700\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMak, M. K. Reduced step length, not step length variability, is central to gait hypokinesia in people with Parkinson's disease. \u003cem\u003eClin. Neurol. Neurosurg.\u003c/em\u003e \u003cb\u003e115\u003c/b\u003e, 587\u0026ndash;590. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.clineuro.2012.07.014\u003c/span\u003e\u003cspan address=\"10.1016/j.clineuro.2012.07.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchniepp, R. et al. Locomotion speed determines gait variability in cerebellar ataxia and vestibular failure. \u003cem\u003eMov. Disord\u003c/em\u003e. \u003cb\u003e27\u003c/b\u003e, 125\u0026ndash;131. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/mds.23978\u003c/span\u003e\u003cspan address=\"10.1002/mds.23978\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRennie, L. et al. The reliability of gait variability measures for individuals with Parkinson's disease and healthy older adults\u0026ndash;The effect of gait speed. \u003cem\u003eGait Posture\u003c/em\u003e. \u003cb\u003e62\u003c/b\u003e, 505\u0026ndash;509. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gaitpost.2018.04.011\u003c/span\u003e\u003cspan address=\"10.1016/j.gaitpost.2018.04.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCole, M. H. et al. Imposed faster and slower walking speeds influence gait stability differently in Parkinson fallers. \u003cem\u003eArch. Phys. Med. Rehabil\u003c/em\u003e. \u003cb\u003e98\u003c/b\u003e, 639\u0026ndash;648 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHartmann, A., Luzi, S., Murer, K., de Bie, R. A. \u0026amp; de Bruin E. D. Concurrent validity of a trunk tri-axial accelerometer system for gait analysis in older adults. \u003cem\u003eGait Posture\u003c/em\u003e. \u003cb\u003e29\u003c/b\u003e, 444\u0026ndash;448 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMonaghan, K., Delahunt, E. \u0026amp; Caulfield, B. Increasing the number of gait trial recordings maximises intra-rater reliability of the CODA motion analysis system. \u003cem\u003eGait Posture\u003c/em\u003e. \u003cb\u003e25\u003c/b\u003e, 303\u0026ndash;315. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gaitpost.2006.04.011\u003c/span\u003e\u003cspan address=\"10.1016/j.gaitpost.2006.04.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVila, M. H., P\u0026eacute;rez, R., Mollinedo, I. \u0026amp; Cancela, J. M. Analysis of gait for disease stage in patients with Parkinson's disease. \u003cem\u003eInt. J. Environ. Res. Public. Health\u003c/em\u003e. \u003cb\u003e18\u003c/b\u003e, 720. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph18020720\u003c/span\u003e\u003cspan address=\"10.3390/ijerph18020720\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCanning, C. G., Ada, L., Johnson, J. J. \u0026amp; McWhirter, S. Walking capacity in mild to moderate Parkinson's disease. \u003cem\u003eArch. Phys. Med. Rehabil\u003c/em\u003e. \u003cb\u003e87\u003c/b\u003e, 371\u0026ndash;375 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChatzaki, C. et al. Can gait features help in differentiating Parkinson's disease medication states and severity levels? A machine learning approach. \u003cem\u003eSensors\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e, 9937 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCurtze, C., Nutt, J. G., Carlson-Kuhta, P., Mancini, M. \u0026amp; Horak, F. B. Levodopa is a double-edged sword for balance and gait in people with Parkinson's disease. \u003cem\u003eMov. Disord\u003c/em\u003e. \u003cb\u003e30\u003c/b\u003e, 1361\u0026ndash;1370 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVabalas, A., Gowen, E., Poliakoff, E. \u0026amp; Casson, A. J. Machine learning algorithm validation with a limited sample size. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e14\u003c/b\u003e, e0224365 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlbrecht, F. et al. Unraveling Parkinson's disease heterogeneity using subtypes based on multimodal data. \u003cem\u003eParkinsonism Relat. Disord\u003c/em\u003e. \u003cb\u003e102\u003c/b\u003e, 19\u0026ndash;29 (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e. Physical and clinical characteristics of all participants.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"626\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eCharacteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003eControls\u003c/p\u003e\n \u003cp\u003e(\u003cem\u003en\u003c/em\u003e = 75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003eEarly PDs\u003c/p\u003e\n \u003cp\u003e(\u003cem\u003en\u003c/em\u003e = 61)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003eModerate PDs\u003c/p\u003e\n \u003cp\u003e(\u003cem\u003en\u003c/em\u003e = 42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eSex (male/female)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e33/42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e36/25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e28/14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.050\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e68.91\u0026plusmn;5.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e68.61\u0026plusmn;6.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e69.45\u0026plusmn;5.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.766\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eHeight (cm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e160.61\u0026plusmn;8.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e161.89\u0026plusmn;8.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e160.15\u0026plusmn;8.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.560\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eBody mass (kg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e63.94\u0026plusmn;9.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e64.09\u0026plusmn;9.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e61.65\u0026plusmn;10.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.397\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e24.75\u0026plusmn;3.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e24.41\u0026plusmn;2.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e23.89\u0026plusmn;2.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.279\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eK-MMSE (scores)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e27.28\u0026plusmn;1.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e27.75\u0026plusmn;1.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e27.6\u0026plusmn;1.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.253\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eHoehn and Yahr scale (stages)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e1.84\u0026plusmn;0.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e2.69\u0026plusmn;0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eUPDRS total (scores)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e52.00\u0026plusmn;16.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e66.81\u0026plusmn;17.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eUPDRS part Ⅲ (scores)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e32.84\u0026plusmn;12.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e42.23\u0026plusmn;13.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003ePIGD (scores)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e0.60\u0026plusmn;0.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e1.17\u0026plusmn;0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eL-Dopa equivalent dose (mg/day)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e571.89\u0026plusmn;330.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e917.48\u0026plusmn;519.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;0.001\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eTreatment duration (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e4.93\u0026plusmn;3.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e6.33\u0026plusmn;3.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.020\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 31.6294%;\"\u003e\n \u003cp\u003eSymptom duration (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e5.36\u0026plusmn;3.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e7.12\u0026plusmn;3.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.0927%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.012\u003c/strong\u003e\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" style=\"width: 100%;\"\u003e\n \u003cp\u003eAll data are indicated as mean \u0026plusmn; standard deviation; PDs: Parkinson\u0026rsquo;s disease; BMI: Body mass index; K-MMSE: Korean mini-mental state examination; UPDRS: Unified Parkinson\u0026rsquo;s disease rating score; PIGD: Postural instability/gait difficulty. \u003csup\u003ea\u003c/sup\u003e Mann\u0026ndash;Whitney U test; \u003csup\u003eb\u003c/sup\u003e Independent sample T-test; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e. Model parameters of seven classifiers estimated by grid search.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eMLs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003ePDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003ePDs\u003c/p\u003e\n \u003cp\u003e(3 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003eEarly PDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003eEarly PDs\u003c/p\u003e\n \u003cp\u003e(2 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eEarly PDs vs.\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;Moderate PDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003eEarly PDs vs.\u003c/p\u003e\n \u003cp\u003eModerate PDs\u003c/p\u003e\n \u003cp\u003e(1 Feature)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003eC\u003c/em\u003e = 0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e\u003cem\u003ek\u003c/em\u003e = 7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eLDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003en_components=1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eQDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param = 0.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param=0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param=0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param = 0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param = 0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003ereg_param = 0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 7.6\u003cbr\u003e\u0026nbsp;gamma = scale,\u003cbr\u003e\u0026nbsp;kernel = linear\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 29.5,\u003cbr\u003e\u0026nbsp;gamma = 0.01,\u003cbr\u003e\u0026nbsp;kernel = rbf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 1.9,\u003cbr\u003e\u0026nbsp;gamma = 0.001,\u003cbr\u003e\u0026nbsp;kernel = rbf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 1719,\u003cbr\u003e\u0026nbsp;gamma = 0.0001,\u003cbr\u003e\u0026nbsp;kernel = rbf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 0.5,\u003cbr\u003e\u0026nbsp;gamma = 1.0,\u003cbr\u003e\u0026nbsp;kernel = rbf\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003eC = 7.6,\u003cbr\u003e\u0026nbsp;gamma = 0.01,\u003cbr\u003e\u0026nbsp;kernel = rbf\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 5%;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=10,\u003cbr\u003e\u0026nbsp;n_estimators=1250\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=30,\u003cbr\u003e\u0026nbsp;n_estimators=1250\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=15,\u003cbr\u003e\u0026nbsp;n_estimators=1500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=15,\u003cbr\u003e\u0026nbsp;n_estimators=500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=30,\u003cbr\u003e\u0026nbsp;n_estimators=750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15%;\"\u003e\n \u003cp\u003emax_depth=10,\u003cbr\u003e\u0026nbsp;n_estimators=500\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"7\" style=\"width: 100%;\"\u003e\n \u003cp\u003eML: Machine learning; PDs: People with Parkinson\u0026rsquo;s disease; LR: Logistic regression, \u0026ldquo;C\u0026rdquo; is the inverse of the regularization strength; KNN: K-nearest neighbor, \u0026ldquo;\u003cem\u003ek\u003c/em\u003e\u0026rdquo; is the number of neighbors; NB: Na\u0026iuml;ve Bayes; LDA: Linear discriminant analysis, \u0026ldquo;n_components\u0026rdquo; is the number of components; QDA: Quadratic discriminant analysis, \u0026ldquo;reg_param\u0026rdquo; is the regularization of the per-class covariance; SVM: Support vector machine, \u0026ldquo;C\u0026rdquo; is the regularization parameter, and \u0026ldquo;gamma\u0026rdquo; is the kernel coefficient; RF: Random forest, \u0026ldquo;n_estimators\u0026rdquo; is the number of trees in the forest, and \u0026ldquo;max_depth\u0026rdquo; is the maximum depth of the tree.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"625\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"9\" style=\"width: 625px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e. Results of binary logistic regression for different groups analyzed.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eEstimate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eSE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eOR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e95% CI for OR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 71px;\"\u003e\n \u003cp\u003eR\u003csub\u003eN\u003c/sub\u003e\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eControls and PDs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"6\" style=\"width: 424px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 5px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eCV of stride length at the LWS (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.599\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e1.820\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e1.101\u0026ndash;3.007\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.020\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" rowspan=\"3\" style=\"width: 71px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.528\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eWalking speed at the PWS (m/s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e\u0026minus;0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.392\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.421\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e0.195\u0026ndash;0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.027\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eStride length at the HWS (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e\u0026minus;1.642\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.460\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e0.079\u0026ndash;0.477\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eControls and early PDs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"6\" style=\"width: 424px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 5px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eCV of stride length at the LWS (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.535\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.240\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e1.707\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e1.067 \u0026ndash; 2.732\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.026\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" rowspan=\"2\" style=\"width: 71px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.404\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eStride length at the HWS (m)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e\u0026minus;1.790\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.364\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.167\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e0.082 \u0026ndash; 0.341\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eEarly PDs and moderate PDs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"6\" style=\"width: 424px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 5px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 196px;\"\u003e\n \u003cp\u003eWalking speed at the PWS (m/s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e\u0026minus;0.979\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.282\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0.376\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 119px;\"\u003e\n \u003cp\u003e0.216 \u0026ndash; 0.653\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 71px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.291\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"9\" style=\"width: 625px;\"\u003e\n \u003cp\u003eDependent variable 1 = Controls, 2 = PDs; 1 = Controls, 2 = H\u0026amp;Y 1\u0026ndash;2; 1 = H\u0026amp;Y 1\u0026ndash;2, 2 = H\u0026amp;Y 2.5\u0026ndash;3.\u0026nbsp;This model adjusted for age, gender, height, BMI, and MMSE score; PDs: People with Parkinson\u0026rsquo;s disease; Early PDs: H\u0026amp;Y 1\u0026ndash;2; Moderate PDs: H\u0026amp;Y 2.5\u0026ndash;3; CV: Coefficient of variance; PWS: Preferred walking speed; LWS: Lower walking speed; HWS: Higher walking speed; SE: Standard error; OR: Odds ratio; CI: Confidence interval; R\u003csub\u003eN\u003c/sub\u003e\u003csup\u003e2\u003c/sup\u003e is the fit statistic for the Nagelkerke model; \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"7\" style=\"width: 602px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 4\u003c/strong\u003e. Accuracies of seven classifiers from five-fold cross-validation.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eMLs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003ePDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003ePDs\u003c/p\u003e\n \u003cp\u003e(3 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003eEarly PDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eControls vs.\u003c/p\u003e\n \u003cp\u003eEarly PDs\u003c/p\u003e\n \u003cp\u003e(2 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003eEarly PDs vs.\u003c/p\u003e\n \u003cp\u003eModerate PDs\u003c/p\u003e\n \u003cp\u003e(30 Features)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003eEarly PDs vs.\u003c/p\u003e\n \u003cp\u003eModerate PDs\u003c/p\u003e\n \u003cp\u003e(1 Feature)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e67.9 \u0026plusmn; 9.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e69.4 \u0026plusmn; 6.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e65.3 \u0026plusmn; 4.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.0 \u0026plusmn; 4.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.8 \u0026plusmn; 14.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e68.1 \u0026plusmn; 9.3\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.6 \u0026plusmn; 6.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e67.9 \u0026plusmn; 6.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e62.7 \u0026plusmn; 5.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.0 \u0026plusmn; 6.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e70.5 \u0026plusmn; 11.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e63.9 \u0026plusmn; 11.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.5 \u0026plusmn; 7.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e68.9 \u0026plusmn; 7.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e63.3 \u0026plusmn; 11.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e67.3 \u0026plusmn; 8.6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e56.7 \u0026plusmn; 12.9\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e69.0 \u0026plusmn; 10.6\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eLDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e65.5 \u0026plusmn; 8.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e68.4 \u0026plusmn; 8.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.7 \u0026plusmn; 9.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.7 \u0026plusmn; 3.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e57.5 \u0026plusmn; 8.7\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e68.1 \u0026plusmn; 9.3\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eQDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e68.9 \u0026plusmn; 6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e69.9 \u0026plusmn; 7.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.7 \u0026plusmn; 5.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.0 \u0026plusmn; 8.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.8 \u0026plusmn; 6.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e69.8 \u0026plusmn; 11.4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.0 \u0026plusmn; 8.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e66.5 \u0026plusmn; 9.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e67.3 \u0026plusmn; 4.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e64.7 \u0026plusmn; 3.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e75.5 \u0026plusmn; 5.4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e69.8 \u0026plusmn; 11.4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 38px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e79.1 \u0026plusmn; 5.4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e78.1 \u0026plusmn; 9.0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e71.3 \u0026plusmn; 6.9\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e61.3 \u0026plusmn; 11.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 94px;\"\u003e\n \u003cp\u003e75.4 \u0026plusmn; 6.4\u003csup\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 94px;\"\u003e\n \u003cp\u003e66.5 \u0026plusmn; 10.2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"7\" style=\"width: 602px;\"\u003e\n \u003cp\u003eMean (%) \u0026plusmn; standard deviation (%) is calculated from the five-fold cross-validation; the mean values presented in boldface denote the best performance (the highest test accuracy); MLs: Machine learning techniques; LR: Logistic regression; KNN: K-nearest neighbors; NB: Naїve Bayes; LDA: Linear discriminant analysis; QDA: Quadratic discriminant analysis; SVM: Support vector machine; RF: Random forest.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 3px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Parkinson’s disease, gait, severity, motor symptom, artificial intelligence, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-5195774/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5195774/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eEarly detection of Parkinson's disease (PD) and accurate assessment of disease progression are critical for optimizing treatment and rehabilitation. However, there is no consensus on how to effectively detect early-stage PD and classify motor symptom severity using gait analysis. This study evaluated the accuracy of machine learning (ML) models in classifying early and moderate stages of PD based on spatiotemporal gait features at different walking speeds. A total of 178 participants were recruited, including 103 individuals with PD (61 early-stage, 42 moderate-stage) and 75 healthy controls. Participants performed a walking test on a 24-meter walkway at three speeds: preferred walking speed (PWS), 20% faster (HWS), and 20% slower (LWS). Key features\u0026mdash;walking speed at PWS, stride length at HWS, and the coefficient of variation (CV) of stride length at LWS\u0026mdash;achieved a classification accuracy of 78.1% using the random forest algorithm. For early PD detection, stride length at HWS and CV at LWS provided 67.3% accuracy with Na\u0026iuml;ve Bayes. Walking speed at PWS was the most critical feature for distinguishing early from moderate PD, with an accuracy of 69.8%. These findings suggest that assessing gait over consecutive steps under different speed conditions may improve early detection and severity assessment of people of PD.\u003c/p\u003e","manuscriptTitle":"Machine Learning for Early Detection and Severity Classification in People with Parkinson's Disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-15 07:37:28","doi":"10.21203/rs.3.rs-5195774/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-21T03:52:25+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-17T23:04:09+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-11T09:47:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"169025274271724812810539243082997330998","date":"2024-10-10T06:28:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"35781289338220967914234851355843529796","date":"2024-10-10T06:23:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-10-10T01:08:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-10-10T01:06:34+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-10-04T14:17:55+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-10-03T05:33:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-10-03T04:32:49+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"38fa7293-76e0-424b-8ec8-49b4201e4631","owner":[],"postedDate":"November 15th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":39186539,"name":"Health sciences/Biomarkers"},{"id":39186540,"name":"Health sciences/Diseases"},{"id":39186541,"name":"Health sciences/Health care"}],"tags":[],"updatedAt":"2025-01-06T15:58:40+00:00","versionOfRecord":{"articleIdentity":"rs-5195774","link":"https://doi.org/10.1038/s41598-024-83975-3","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-01-02 15:56:57","publishedOnDateReadable":"January 2nd, 2025"},"versionCreatedAt":"2024-11-15 07:37:28","video":"","vorDoi":"10.1038/s41598-024-83975-3","vorDoiUrl":"https://doi.org/10.1038/s41598-024-83975-3","workflowStages":[]},"version":"v1","identity":"rs-5195774","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5195774","identity":"rs-5195774","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}