Prediction and visualization of causes-of-deaths seasonality and death excess rates using machine-learning approaches

Besides a social-psychological impact on individuals, death as a phenomenon is an important event of interest in various statistical applications and modeling, particularly in demography predictions or survival analysis. While some aspects of death’s statistical properties such as rough frequencies based on different causes-of-deaths in times are known, some others are under current cutting-edge research, namely (i) the death seasonality, which means that causes-of-deaths have significantly different time periods when emerge and, thus, increase death rates asynchronously; and (ii) the death excess, which stands for a situation when the death rate is higher than expected. Furthermore, considering COVID-19 as a worldwide emergency issue, closely related to deaths distributions and their changes both in space and time, which is indisputable a game-changer, the prediction of both the death seasonality and death excess became even more tricky. The precise death rates, death seasonality, and death excess predictions may have indisputable application in many areas, especially in demography and actuary for tuning and better performance of predictive models for insurance distributions, and also in all statistical modeling considering death rates and other aspects as dependent or independent variables, e. g. longitudinal models, survival analysis models, and lastly but not lastly, time series.
Consequently, better predictions of the death seasonality and excess could enable better healthcare management and early and targeted healthcare system preparation and reaction when needed, particularly in the case of COVID-19 as a death cause. In this project, we improve and build death seasonality and death excess predictive models, using traditional approaches such as nonparametric smoothing and machine-learning techniques. Data we use for model-learning describe on an individual level the causes-of-deaths in time and other social-economic covariates. Besides the predictions, we also work on effective visualization of the death seasonality and death excess for various causes-of-deaths in time, using R shiny implementation into a clickable web-based application.