Table 3 Big data modeling techniques

Neural Networks

Discover non-linear relationships. Can assess multi-level interactions

“Black Box” to clinicians; hard to implement into a DSS*

Random Forests

Finds most probable solution set; robust against scaling influences

Not always best in terms of prediction; hard to implement into a DSS

Cluster Analysis

Finds groups of very similar patients; exploratory analysis

Unsupervised technique

Principal Components Analysis

Uncovers the variables contributing the most to outcome variation

Not amenable to binary outcomes; assumes additive relationship

Support Vector Machines

Robust against statistical assumptions

Difficult to implement into a DSS