Statistical Considerations on the Design Sample Clauses

Statistical Considerations on the Design. Preliminary correlations between heterogeneous information sets related to resilience will be extracted and hypotheses to be used as input to the model will be defined. Biological, clinical treatment, psychological, lifestyle, social and environmental data will be considered. Concrete hypotheses will be formulated based on the correlations to be extracted. To this end both retrospective data from the BOUNCE clinical partners and literature information will be exploited. A broad palette of statistical and machine learning techniques will be used in order to identify risk factors for poor resilience and hidden correlations among data. A methodological approach is described below. • Univariate (e.g. t-test, chi-squared test, Xxxx–Xxxxxxx U test, Xxxxxxxx'x rank correlation coefficient etc.) and multivariate techniques (e.g. logistic regression, correlation-based feature selection, sequential forward selection, sequential backward elimination, decision trees, naive Bayes etc.) will be performed to identify features of importance. • Intelligent pattern recognition analysis of an individual’s context will be applied to allow the identification of established behaviours and, eventually, cause and effect relationships. • Given the sequential nature of recorded data, association analysis techniques, able to handle both co- occurrence and dynamic relationships in multivariate time series data, will be utilized. • Temporal data mining will enable the identification of dynamic patterns or predictive rules in long-term trajectories and, eventually, will allow drawing conclusions regarding the associations between the patient’s context - indicating resilience to BC - and the clinical health outcomes, and vice versa. A brief presentation of this approach is provided in chapter 8 [M].