What's the other major approach/paradigms in machine learning besides Baysian methods? [closed]

Question

Closed. This question needs to be more focused. It is not currently accepting answers.

---

Want to improve this question? Update the question so it focuses on one problem only by editing this post.

Closed 8 years ago.

Improve this question

I just started my journey into the Machine Learning field. So far I know that Bayesian method is one of the major approaches in this field. So what other options are there? And any comparison between them, such as which approach should be used where.

Answer 1

1. Decisions trees is a general machine learning approach for building trees that are used for classification. Each intermediate node of the tree has a condition associated with it that is used to select right or left sub-tree for future classification. Leaves of a tree represent classes of the problem. Take a look at such algorithms as ID3 or C4.5 (improvement of ID3). Decision trees are fast and can be used for creation of non-linear decision boundaries.

2. Artificial neural network is a classifier that contains several layers of neurons. Each neuron is function that has multiple inputs and a single output. Inputs of a neuron are outputs of other neurons or inputs of the classifier (features). Signals in neural networks usually go in one directions from inputs to output, but recursive neural networks have also been investigated. The most important algorithm for neural networks training is back propagation algorithm - a famous algorithm that is used for training multilayer neural networks.

3. Meta-classifiers or ensemble classifiers is method of combining different classifiers to perform a single classification. The simplest method is voting, but there are more advanced methods such as: stacking, grading, boosting, AdaBoost, Random Forest, Rotation Forest, etc.

4. Logistic regression is another popular machine learning algorithm. It can only create linear decision boundary, but should work faster than other more sophisticated algorithms and can be a good selection for a first algorithm to try when you face a new problem.

5. SVM is one of the most commonly used algorithms in the field of machine-learning. It seems that it was used for most of classification problems from text classification to image classification. This algorithm is based on non-linear transformation of the dataset so that linear decision boundary could be created for classification.

6. Regression analysis unlike classification algorithms is used for prediction of continuous values (such as house prices) from features values. If you would like to investigate this branch of machine learning you should start with Linear Regression algorithm.

7. In contrast to supervised learning where for each example you have a target value, unsupervised learning approach is used for analysis of unlabeled data, such a set of articles. Unsupervised learning is used not for prediction or classification, but for detecting groups of related items. You should probably start with k-means clustering if you are going to learn this topic.

8. Reinforcement learning is an approach when the data is presented to a learning agent when the agent reaches some result. For example an agent that learns how to play chess is "rewarded" if it wins a game and "punished" if he loses but receives no useful information about results of its intermediate steps. Algorithms of reinforcement learning are used to deduct optimal strategy of actions when no information about optimal action in each particular situation is available.