Pathological and age-related changes may affect an individual's gait, in turn raising the risk of falls. In elderly, falls are common and may eventuate in severe injuries, long-term disabilities, and even death. Thus, there is interest in estimating the risk of falls from gait analysis. Estimation of the risk of falls requires consideration of the longitudinal evolution of different variables derived from human gait. Bayesian networks are probabilistic models which graphically express dependencies among variables. Dynamic Bayesian networks (DBNs) are a type of BN adequate for modeling the dynamics of the statistical dependencies in a set of variables. In this work, a DBN model incorporates gait derived variables to predict the risk of falls in elderly within 6 months subsequent to gait assessment. Two DBNs were developed; the first (DBN1; expert-guided) was built using gait variables identified by domain experts, whereas the second (DBN2; strictly computational) was constructed utilizing gait variables picked out by a feature selection algorithm. The effectiveness of the second model to predict falls in the 6 months following assessment is 72.22%. These results are encouraging and supply evidence regarding the usefulness of dynamic probabilistic models in the prediction of falls from pathological gait.

Abstract. In many classification problems, and in particular in medical domains, it is common to have an unbalanced class distribution. This pose problems to classifiers as they tend to perform poorly in the minority class which is often the class of interest. One commonly used strategy that to improve the classification performance is to select a subset of relevant features. Feature selection algorithms, however, have not been designed to favour the classification performance of the minority class. In this paper, we present a novel filter feature selection algorithm, called FSMC, for unbalanced data sets. FSMC selects attributes that have minority class distributions significantly different from the majority class distributions. FSMC is fast, simple, selects a small number of features and outperforms in most cases other feature selection algorithms in terms of global accuracy and in terms of performance measures for the minority class such as precision, recall, F-measure and ROC values.