Transformational MTS

Transformational MTS is now available on Amazon.

Since I first learned of the Mahalanobis-Taguchi System, I was fascinated by its potential. I continue to be amazed that the Mahalanobis Distance was introduced in 1936, long before computers could help in the calculations. For years I sought to combine the power of computing with MTS to enhance the capability.

My first efforts were simply applying full factorial analysis to datasets, rather than the orthogonal arrays that Taguchi used prior to widespread computer use. While there was sometimes a modest improvement discovered, memory limitations prevented the analysis of more than 21 or so factors. Sometimes it increased the SN ratio, but without really improving overall discrimination. That the improvement was only modest using full factorial analysis is a testament to the power of using orthogonal arrays to optimize the datasets. However, it marked the beginning of a six year journey for an improved algorithm that was capable of starting with a dataset of at least 200 factors.

I eventually developed an algorithm that not only allowed for datasets to be quickly iterated and optimized, but provided evidence that datasets could be transformed to make patterns more clearly recognized in the multi-dimensional (but linear) Mahalanobis space. It seems fitting that the improved algorithm is not one that can be fully automated, but still requires human assessment and expertise to implement. It is computer aided, but human powered.

For some datasets, this new insight can lead directly to function definitions. Whether better discrimination is found in Mahalanobis space or a defined function depends on how many factors are essential and the linearity of the system. This creates two possible outcomes for an MTS project. One possible outcome is a definition of Mahalanobis space that can calculate a useful Mahalanobis Distance to determine how likely a data point is to be a part of the normal population. The other possible outcome is that insight gained through the MTS analysis leads to data transformations that enable a function to be defined, enabling faster real time analysis of unknown samples.