Data Science and Predictive Analytics

Abstract

Since the turn of the twenty-first century, the evidence overwhelming reveals that the rate of increase for the amount of data we collect doubles each 12–14 months (Kryder’s law). The growth momentum of the volume and complexity of digital information we gather far outpaces the corresponding increase of computational power, which doubles each 18 months (Moore’s law). There is a substantial imbalance between the increase of data inflow and the corresponding computational infrastructure intended to process that data. This calls into question our ability to extract valuable information and actionable knowledge from the mountains of digital information we collect. Nowadays, it is very common for researchers to work with petabytes (PB) of data, 1PB ¼ 1015 bytes, which may include nonhomologous records that demand unconventional analytics. For comparison, the Milky Way Galaxy has approximately 2 1011 stars. If each star represents a byte, then one petabyte of data correspond to 5,000 Milky Way Galaxies. This data storage-computing asymmetry leads to an explosion of innovative data science methods and disruptive computational technologies that show promise to provide effective (semi-intelligent) decision support systems. Designing, understanding and validating such new techniques require deep within-discipline basic science knowledge, transdisciplinary team-based scientific collaboration, openscientific endeavors, and a blend of exploratory and confirmatory scientific discovery. There is a pressing demand to bridge the widening gaps between the needs and skills of practicing data scientists, advanced techniques introduced by theoreticians, algorithms invented by computational scientists, models constructed by biosocial investigators, network products and Internet of Things (IoT) services engineered by software architects.