Short answers, pulled from the story.
Big data refers to datasets too large or complex to be handled by traditional data-processing software, where parallel computing tools are required. The term has been in use since the 1990s, with John Mashey credited with popularizing it. A 2018 formal definition notes that big data represents a distinct change in computer science, moving away from the guarantees of Codd's relational model toward parallel programming.
Big data is commonly described by volume (quantity of data), variety (types of data, including structured and unstructured), velocity (speed of data generation and processing), and veracity (reliability and quality of the data). Later frameworks added value (the worth of derived insights) and variability (shifting formats and sources). Researchers Kitchin and McArdle found that none of these characteristics appear consistently across all real-world cases.
An estimated 2.5 exabytes of data are generated every day. IDC projected that global data volume would grow from 4.4 zettabytes in 2013 to 44 zettabytes by 2020, and to 163 zettabytes by 2025. The world's technological capacity to store information has roughly doubled every 40 months since the 1980s.
Key technologies include Google's MapReduce framework (described in a 2004 paper), Apache Hadoop (an open-source implementation of MapReduce), and Apache Spark (developed in 2012 to add in-memory processing). Teradata's parallel-processing DBC 1012 system, marketed in 1984, was among the earliest commercial big data platforms. The HPCC Systems platform, built by Seisint Inc. in 2000 and open-sourced in 2011, uses a declarative dataflow language called ECL.
Critics argue that big data carries a false aura of objectivity and accuracy. Google Flu Trends overstated flu outbreaks by a factor of two. Researchers Reips and Matzat called big data a scientific fad in 2014. Catherine Tucker wrote that big data is unlikely to be valuable by itself, with processing skills mattering more than raw volume. Large datasets are also prone to spurious correlations and the multiple comparisons problem, where testing many hypotheses simultaneously produces false positives.
Big data analytics in healthcare supports personalized medicine, clinical risk prediction, automated patient reporting, and computer-aided diagnosis. Epilepsy monitoring alone generates five to ten gigabytes of data per day per patient, and a single uncompressed breast tomosynthesis image averages 450 megabytes. McKinsey estimated in 2011 that effective big data use in U.S. healthcare could generate more than $300 billion in value annually, though the sector faces significant challenges around data accuracy and patient privacy.