Using basic machine learning algorithms to classify mobile phone data-Naive Bayes

1 Algorithm ideas

The specific ideas of Naive Bayes will not be elaborated, and the emphasis will be on how to train a reliable Naive Bayes classifier from existing data.

First, let’s take a look at the data csv table that was preprocessed last time. Each row of the table is the data of a mobile phone model, each column is an attribute of the mobile phone, and the last column is the label value, with only high and low labels.

According to Bayesian formula:

Our goal is that the classifier can be based on the input data $\text{x}(x_1,x_2,...,x_n)$ Find the most likely$C_i$. Because of the denominator $p\left(\text{x}\right)$ is fixed, so we only need to compare the value of each numerator.

In the numerator:

• $P(C_i)$ Is the probability of the category, just divide the number of data under the category by the total number of data.
• $p(\text{x}|C_i)$ Is$C_i$X (x 1, x 2,..., Xn) \textbf(x)(x_1,x_2,...,x_n) appears in$\text{x}(x_1,x_2,...,x_n)$ . Because we are a naive Bayes algorithm, all attributes are directly not related by default, so:$p(\text{x}|C_i)$ = $p(x_1|C_i)\ast ...\ast p(x_n|C_i)$

In the data table above, the data is continuous rather than discrete. How to know $p(x|C_i)$ Value?

Here, the data can be divided into k categories according to intervals, so as to convert continuous data into discrete data. So that each $x_i$There are k kinds of values, and their probabilities in the training set can be calculated one by one.

So far, we must distinguish the various concepts in Naive Bayes:

• Category: The category and label to which the data belongs is the target we want to classify
• Attribute: A piece of data contains multiple different attributes, and each attribute independently affects its category
• Property value: the value of the property

2 code implementation