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

Variance is how to spread out our predicted data points are. Usually, the
variance is a result of overfitting to the sample data we used to create the
model. It doesn't do very well at predicting the outcome of new variables.
There will always be some level of error in your models. It’s a fact of life
that no matter how good you are at predicting something, there is always
some random or nonrandom variation in the universe that will make your
prediction slightly off from the true outcome.
I’ve created a visual example of four bullseye targets to help illustrate the
difference between models suffering from high bias and variance. In this
instance, the center of the bullseye represents the true value that our model
is trying to predict. The top left corner is the ideal model. Notice that all our
predicted data points are falling right on the bullseye. This model is quite
accurate and places our predicted data points all around the true value. This
is because of low variance; a lack of ‘spread out’ data points, and low bias;
underfitting that skews our results.

---

In the top right target, the model is suffering from high variance. You can
see that our data points are clustered around the bullseye. Unfortunately, the
average distance between the predicted values and the bullseye is high due
to high variance.
In the bottom left target, the model didn’t suffer much from high variance.
The average distance between the predicted data points is low, but they
aren’t clustered around the bullseye but slightly off it as a result of high
bias. This is probably the result of too little training data, which means that
the model doesn’t perform well when it gets introduced to new data.
The bottom right model suffers from both high variance and high bias. In
this worst-case scenario, the model is very inaccurate because the average
distance between predicted data points and the true value is high, and the
predicted data points are skewed.
Variance can be caused by a significant degree of correlation between
variables. If you use too many independent variables, this can also be a
cause of the high variance. Sometimes, if the variance is too high, we can
combat that by allowing a small amount of bias in the model. This is known
as regularization. We’ll cover that a little later.
In statistics, the population is the group of people or the set of data you are
trying to analyze. The sample is the subgroup of that population, whose
data you use to create your model. The parameters are the characteristic of
the variables of the population that you are trying to identify and make
predictions from in your model.

---

Descriptive statistics is the use of data to examine a population. Typically,
descriptive statistics involve the mean or average, mode, media, size,
correlation. Machine learning falls into the category of inferential statistics
because we are using the data to find patterns and relations but also to make
predictions based on this information. Inferential statistics, or descriptive
stats, is using the characteristics of your population to make predictions.
This is where your regression models and classification models will come
in. When we infer something, we make a logical deduction about a
population-based and the knowledge we are given.
When you are looking at data, you should also take note of the