Relevance of Simple Predictive Models

Simple Predictive Models:

Despite the explosion of interest in complex predictive models such as SVN (Subversion, i.e. version control system), random forests and deep learning, much benefit may be derived from relatively simple models such as linear or logistic regression and simple (bivariate) curve-fitting. Simple models are easy to construct (linear models have been a part of spreadsheet software such as Excel for years), explain, and justify.Further, not every business problem requires the enormous complexity of thousands of degrees of freedom. In the age of so-called "big data", even very large organizations may be faced with data sets which are actually quite small. Consider data gathered daily over a few years: the count of daily observations may be in the hundreds or even low thousands, but the number of times a complete cycle of annual seasonality may still be in the single digits!Additionally, simple models are much easier to deploy than complex ones: They are easier to verify and audit, and typically require no auxiliary software (libraries, APIs [application programming interface], computer servers, etc.). Many are small enough to fit into a single spreadsheet cell or run on a programmable calculator.Important organizational questions are still faced by managers which require better answers than guessing or human hunches, but which do not present sufficient data for large-scale data science solutions. Readers should consider linear regression, logistic regression, statistical summaries, and their associated confidence intervals when approaching new problems.Reference: "Introduction to Statistical Analysis" by Dixon and Massey