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

Supervised learning
In supervised learning, programmers use labeled data. Before we begin
using the algorithms, the data that we are looking at is already
predetermined. We know the inputs and outputs we are looking for. X, and
Y. We are trying to find a relationship between X and Y that we have
chosen.

---

After you find a relationship between X and Y, you get a model, which will
predict an outcome based on those relationships that your machine has
observed in the data. Supervised learning is used for regression and
classification models. In machine learning, we refer to features as a certain
measurable property or characteristic of the data.
The first type of supervised learning that we'll talk about and the first type
of statistical model is called a regression. Regression is a model where the
data input and output are continuous. There are different types of
regression, but the most basic form is linear regression. We use linear
regression to find a relationship between some input X and an output Y.
Once we have estimated this relationship, we can predict Y with X. Linear
models can, and usually do, have more than one X. In regression; output Y
has a numerical value.
Regression Analysis
Regression is the simplest type of machine learning; this is usually where
you start when you are first learning how to use your data. You have a set of
X values, and you want to study their relationship with Y, the output. Our
independent variables, the Xs in our model, are given weight, and for each
value of X, it is multiplied by weight until the aggregate function creates a
prediction for Y.
We can create a predictive model for Y by using data where we already
know the X and Y. Regressing this information will give us the weights of
X. If we have enough relevant data, eventually we will be able to predict
and unknown Y with known values for X.

---

We graph our known Y and X values on a scatterplot, and our regression
model finds the “best fit” line through the data points. The regression line is
called a hyperplane. The steepness of the line is called the slope.
We can measure the distance between the predicted value and actual data
point, and we call that measurement deviation. Our goal when we create a
linear regression is the minimize the deviation in our predictions. The
smaller that difference, the deviation is, the more accurate your model is.
Most of the statistical models used in machine learning are rooted in this
first algorithm. Creating a model that will predict an outcome by plotting
our data points along a line or in clusters. But the line isn’t always straight,
and sometimes the line doesn’t show us the best fit.
An example of a regression function that is nonlinear is the Sigmoid
function. The Sigmoid function creates an S-shaped curve. Instead of
predicting a value, the Sigmoid function takes independent variables and
produces a probability between one and zero.