Inquidia's accomplished in the modern statistical/machine learning techniques that support Data Analytics and Data Science. We see three overlapping branches of methods commonly deployed by our customers: Parametric Linear Models with Experimental Designs; Forecasting/Time Series; and Machine Learning. Our consultants can show you how to put these techniques to work.
Parametric Linear Models with Experimental Designs
Inquidia distinguishes two fundamental statistical use cases of our customers. Some are just looking for the best predictions of outcomes of interest with little concern for cause and effect, while for others, being able to “attribute” effects to particular factors is paramount. Inquidia's point of departure for the latter work is classical linear and logistic regression superimposed on tight experimental designs. These models are often relevant for the performance measurement demands of Data Analytics.
Inquidia’s adept at many popular forecasting methods, including the generalized exponential smoothing framework that estimates trend, seasonality and error (ETS) from time series data. Smoothing often competes with well-traveled autoregressive moving average models (ARIMA), classical regression with time components and supervised machine learning models such as MARS, and Additive Regression.
Where there's less concern with cause and effect and more with predictive accuracy, machine learning techniques are often suitable. We start by acknowledging the world isn’t linear by adding cubic splines and interaction effects to our models. And we test for overfitting by vigorous cross validation. We acknowledge that models built with training data are credulous by penalizing coefficient estimates using established “regularization” methods such as Lars, Lasso and Elastic-Net. And we understand the predictive benefits of the wisdom of crowds using ensemble learning methods such as Random Forests and Gradient Boosting. ML is central to the exploration and prediction of Data Science.
Our statistical work isn’t executed in a vacuum. Models and their predictions must often be incorporated into operations through Data Analytics/Data Science applications. Our DA integration services merge R, SAS/WPS and Python-Sci-Kit models/predictions into existing Analytics apps. And we're very bullish on the emerging Spark ecosystem for big data statistics on Hadoop. WPS's Proc Hadoop provides the capability to seamlessly access data from the Hadoop ecosystem with basic SAS and R syntax. Finally, we collaborate with R vendor Revolution Analytics to promote and showcase their enterprise server, RevoDeployR.