Answers are grounded in this paper and cite their sources.
Machine learning algorithms predict headache occurrences from gait data, indicating potential for remote health monitoring.
The use of a smartphone to monitor human health has been of interest for more than a decade. However, currently its main functionality is reduced to tracking human movements (mainly gait) and evaluating statistical indicators. Using a smartphone to predict changes in the state of health by gait is a little-studied area. Existing research in this area is currently limited only to diagnostic tasks, which are solved based on machine learning methods in classifying statistical data on the gait of healthy and sick people. The aim of the work is to study the effectiveness of machine learning algorithms for predicting headaches based on the gait accelerogram in an automated medical decision support system based on measurements by a mobile device. The paper describes a technique for collecting and preprocessing gait data based on the use of a mobile phone. The sample size for the study included 398 examples of accelerograms reflecting the double steps of one of the volunteers (314 for training and 84 for testing). The sample consisted of measurements that preceded the measurements (3–4 hours) with a note in the questionnaire about head pain and measurements in the absence of any changes in the well-being of the volunteer. The sample was balanced by the number of measurements in each class. To improve the quality of the selection of informative sections of the accelerogram, the obtained gait data were processed by the author's algorithm. To predict the occurrence of headaches, the following classifiers were used in the implementation of the Scikit-learn package: support vectors, gradient boosting, adaptive boosting, bagging, random forest, voting regressor and multilayer perceptron (MLP). The best results were shown by the classifiers Random Forest (Accuracy = 0.8) and Adaptive boosting (Accuracy = 0.78). Next, predictions were made based on the direct distribution network from the Keras package of the TensorFlow environment. The prediction was carried out in several steps. When using ensembles of neural networks from 7 and more, the Accuracy value is 0.97. The results of a study of the effectiveness of various machine learning algorithms for predicting headaches based on the analysis of gait accelerograms are presented. It is revealed that the most accurate predictions can be obtained using ensembles of neural networks. The results obtained during the study, using the example of predicting headaches, show the possibility of conducting remote health monitoring by analyzing gait data using a mobile phone in real conditions. The application of the proposed algorithm in practice made it possible to increase the accuracy of predicting headaches according to the gait accelerogram to 0.97.
No takes yet. Share an insight, caveat, or question.
Dorofeev et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: