Background: Multiple sclerosis (MS) is the most common chronic inflammatory disease of the central nervous system. The progressive impairment of gait is one of the most important pathognomic symptoms associated with falls and fear of falling in people with MS (pwMS). 60% of PwMS have a fear of falling, which leads to restrictions in mobility and physical activity and ultimately reduces their overall quality of life. In particular, early identification of fear of falling is crucial as it enables the early implementation of rehabilitation strategies and clinical decision-making to reduce progression. Therefore, the qualitative and quantitative evaluation of gait is an essential aspect of disease management and can provide valuable insights for personalised treatment decisions in pwMS. Objective(s): To identify the most appropriate method from clinical gait analysis to detect fear of falling in pwMS and to identify optimal machine learning (ML) algorithms to predict fear of falling using the complex multidimensional data from gait analysis. Method(s): A multidimensional gait analysis with pressure and motion sensors as well as patient-reported outcomes (PROs) was performed by 1240 pwMS at the MS Centre of the University Hospital Dresden according to a standardised protocol between November 2020 and September 2021. To identify fear of falling, a feature selection ensemble (FS ensemble) was developed and the classification performance of the four ML models Gaussian Naive Bayes (GNB), Decision Tree, k-Nearest Neighbour (kNN) and Support Vector Machines (SVM) was determined and compared using the gait data. The FS ensemble consisted of four filtering methods: Chi-square test, Information gain, Minimum redundancy Maximum relevance and ReliefF. The filter methods produced a ranked list of features. A threshold reduced this list to a subset. A combination method combines these subsets into the final data set for classification. Different thresholds (50%, 25% and 10%) and combination methods (Union, Union 2, Union 3 and Intersection) were analysed. Result(s): The analyses showed that 37% of the 1240 pwMS had a fear of falling (N = 458; age: 51 +/- 16 years, 76.0% women, median EDSS: 4.0). The FS ensemble improved classification performance in most cases. The SVM showed the best performance (F1 score) of the four classification models in the detection of fear of falling using different data sets. The PROs showed the best F1 scores (Early Mobility Impairment Questionnaire F1 = 0.81 +/- 0.00 and 12-Item Multiple Sclerosis Scale F1 = 0.80 +/- 0.00). In the sensor-based methods, the measurements with cognitive dual-task tasks showed poorer F1 scores (F1 = 0.65 +/- 0.01 and F1 = 0.65 +/- 0.00) than the single-task measurements (F1 = 0.69 +/- 0.00 and F1 = 0.69 +/- 0.01). Conclusion(s): Fear of falling is an important psychological risk factor that is associated with an increased risk of falls. In order to integrate a functioning early warning system for fall detection into MS management and follow-up monitoring, it is necessary to detect the relevant gait parameters and assessment methods. ML strategies enable the integration of gait parameters from clinical routine in order to support the initiation of early rehabilitation measures and the adaptation of the disease course modifying therapeutics.Copyright © 2023