Concept of naive Bayes

Introduction to machine learning —— Naive Bayes

Naive Bayes

Naive Bayes （Naive Bayes,NB）： It is a classification algorithm based on probability theory , Based on Bayesian theory , Classify by calculating the probability that the samples belong to different categories , It is a classic classification algorithm .

Bayesian theory ： Based on the best evidence available （ Observe 、 Data and information, etc ）, To calculate the degree of belief （ Or hypothesis 、 opinion 、 proposition ） Effective method . The degree of belief is the confidence in the authenticity and correctness of things .

simple ： pure 、 Rough , It is simple and crude to assume that a given target value is conditionally independent of each other

Naive Bayesian model

Yes m Samples , Each sample has n Features , Output is k Categories , A priori probability is obtained through sample learning （ How many records are there for each category ）, The conditional probability is obtained through sample learning （ In a certain category , Sample present x What's the probability of ）, Through the above learning results , You can calculate X、Y The joint probability distribution of

Laplacian smoothing

Laplacian smoothing ： In the process of classification , Sometimes we encounter the problem of zero probability , That is, the value of an attribute in a sample with prediction does not appear in the sample , The calculation result of the whole probability will be 0. Laplace smoothes through the molecules 、 Adding adjustment to the denominator can solve this kind of problem .

Advantages and disadvantages of naive Bayesian algorithm

NB advantage ：

• Endorsement based on statistics , Classification efficiency is stable
• Support multi category tasks
• Not sensitive to missing data
• Method is simple , The model is easy to explain
• A small amount of calculation , Support massive data
• Support incremental calculation , It can be used as an online forecast

NB shortcoming ：

• A priori probability is required , Different values affect the results
• There is an error rate in classification decision
• Sensitive to the expression of input data
• ” simple “ Your assumptions have a big impact on the results