Machine Learning: 2 Books in 1: Machine Learning for Beginners, Machine Learning Mathematics. An Introduction Guide to Understand Data Science Through the Business Application

Predictions Accuracy and Model Interpretability
Some of the many methods used to learn from statistical data are less
adaptable and extremely restrictive. When "inference" is the target, the use
of easy and comparatively inflexible techniques of statistical learning has
significant benefits. On the other hand, if the target is the generation of
forecasts and predictions flexible models are preferred.
The performance of the model can be estimated based on its accuracy to
predict the occurrence of an event on new input data. A more accurate
model is deemed as a more valuable model. Interpretability of the model
offers insight into the input-output relationship. An interpreted model can
provide insight into the capability of independent features to
generate predictions for the dependent attribute. The problem occurs

---

because, at the expense of interpretability, as model accuracy improves so
does the complexity of the model.
A more accurate model can offer a business more possibilities, advantages,
time, or money. But the model accuracy needs to be optimized for
such prediction. The optimization of accuracy extends the complexity of the
model even further by introducing additional model parameters (and
resources needed to adjust those parameters). It is much easier and
quicker to interpret a model with a relatively small number of parameters.
An input coefficient and an intercept term are part of a linear regression
model. For instance, every single term can be explored to assess how it
contributes to the production of the output. Switching to a logistic
regression model provides greater authority In the context of the
relationships underlying the potential transformation of a function to output,
that too should be explored along with the coefficients.
It is relatively easy to understand a decision tree of small size, but a heavily
loaded decision tree needs a distinct perspective to understand why
the event is predicted to occur. Furthermore, the optimized combination of
several models into one prediction tends to have no significant or timely
interpretation. Interpretation is deemed ancillary to model accuracy.
For example, models designed to separate and classify “spam” emails from
“non-spam” emails as well as models designed to evaluate the price of a
real estate.

---