Non-linear prediction over time-based big data application

The Big Data are increasing exponential every year so that data became very complex and difficult to be processed. To resolve this problem, data management and analysis offer opportunities to improve decisions in critical development areas such as: meteorology, medicine, finance, sociology or internet. But, classical statistics programs encounter their limits in processing large data-sets, so that introduction of such programs in non-sql database applications is required. Existing large-scale processing data-sets frameworks does not provide statistics tools to reduce the complexity of the large data-sets to meaningful results. More, nowadays statistics have meanings in context of predictions, forecasting and estimation requiring non-linear regressions to define the complex equations of such systems. Non-linear regressions offer the best solution for our complex time-series application where observational data are modeled by non-linear functions and multiple independent variables. Our analytic application is based on data came from BTWord serial application, that collected public trackers to obtain information about the performance, scalability and reliability of BitTorrent. We show how descriptive, inductive and non-linear regression statistics may be integrated in our map-reduce application to generate statistics about evolution in time of BitTorrent network.