Objective Evaluation of Freezing of Gait in Patients with Parkinson's Disease through Machine Learning Approaches

Wearable-Sensor-Based Detection and Prediction of Freezing of Gait in Parkinson’s Disease: A Review

“…To further improve FOG detection performance, multiple features and machine-learning (ML) techniques have been used, such as neural networks [36,38,55,66,76,80,85,86,88,89,91], decision trees [25,39,42,45,52,54,58,85], random forests, [39,42,43] naïve Bayes [42,43], nearest neighbor [42], and support vector machines [64,71,74,75,81,83,86]. In addition, anomaly detection [20] and unsupervised machine learning have been attempted [87], but not extensively explored.…”

“…If data points are not easily separable, a kernel can transform the data into a dimension that is linearly separable [125]. SVM for FOG detection achieved 74.7%–99.73% sensitivity and 79.0%–100% specificity [64,71,74,75,81,83,86].…”

“…NN have been frequently used in FOG detection and prediction studies. For FOG detection, model performance achieved 72.2%–99.83% sensitivity and 48.4%–99.96% specificity [36,38,55,66,76,80,85,86,88,89,91]. Neural networks for FOG prediction tended to perform slightly worse, up to 86% sensitivity, 80.25% specificity, and 89% precision [96,97,99].…”

“…Since freezing manifests differently for each person, person-specific models outperformed person-independent models [42,58,74,86] (with some exceptions as in [53]). However, in practice, it is difficult to obtain enough data to develop a model for an individual.…”

“…When evaluating a classification system, ideally, different data are used for training and testing, as in [25,38,51,55,56,64,66,67,85,96,97,99,100], in order to prevent model performance overestimation that can occur when the model is evaluated using data used in model training. Cross-validation is often used when the dataset size is limited, as done in [24,31,32,33,39,42,43,45,52,54,58,74,75,81,86,87,88,89,90,98]. For FOG research, leave-one-person-out cross-validation was the most common.…”