Machine Learning: A Bird's Eye View

Arthur Samuel (1959): "Field of study that gives computers the ability to learn without being explicitly programmed."

Tom Mitchel (1997): "A computer program is said to learn if its performance at a task T, as measured by a performance P, improves with experience E."

Machine learning algorithms & the confusion that follows:

Choosing the best machine-learning algorithm is dependent on several factors, including the size, quality, and nature of the data. Choosing the best algorithm combines business necessity, specification, experimentation, and available time. In this section, we will look at various machine learning algorithms.

Machine Learning algorithms are classified into four types:

• Supervised Learning
• Semi-supervised Learning
• Unsupervised Learning
• Reinforcement Learning

Supervised Learning:

We gave a known dataset with inputs and expected outputs in supervised learning. The machine discovers a way to determine the outcomes given a set of information.

The following are examples of Supervised Learning Algorithms:

A. Classification: Based on the observed values, this machine learning algorithm will determine which group the new observation belongs to.

The following are the various classification algorithms:

1. Logistic Regression: Logistic Regression is a technique for predicting the likelihood of a target variable. Because the nature of the target or dependent variable is dichotomous, there are only two viable classes (0 or 1).

Logistic regression of the following types:

a. Binary or Binomial: The dependent variable will only have two possible values (0 or 1).

b. Multinomial: The dependent variable can contain three or more unordered types or types with no quantitative significance. For instance, "Type A," "Type B," or "Type C."

c. Ordinal: The dependent variable can have three or more ordered types or types with quantitative significance. T-shirt sizes, for example, "small," "medium," "large," "extra-large," and so on.

Image Source

2. Naive Bayes Classifier Algorithm: The Naive Bayes classifier algorithm is based on Bayes' theorem and classifies each value as independent of all others. It enables us to use probability to forecast a category based on a given set of features.

3. Support Vector Machine: SVM effectively categorizes data by supplying training examples, each set of which is labelled as belonging to one of two categories. The programme then constructs a model that allocates new values to one or both categories.

4. Decision Trees: Decision Trees can be used to address problems involving both regression and classification. It is a tree structure that looks like a flow chart and uses a branching mechanism to show every possible outcome of an action. Each node in the tree represents a test on a particular variable, and each branch indicates the result of that test.

5. Random Forests: This ensemble learning method combines numerous algorithms to produce superior classification, regression, and other outcomes.

6. Nearest Neighbors: The K Nearest Neighbor algorithm predicts how likely a data item belongs to one of two groups. It examines the data points surrounding a particular data point to determine which group it belongs to.

B. Regression Analysis: Regression Analysis is a set of machine learning approaches that allows us to predict a continuous result of a variable based on the value of one or more predictor variables.

C. Forecasting: We make future forecasts based on data from the past and present.

• Semi-supervised learning: Semi-supervised learning is a method of machine learning in which a small amount of labelled data is combined with a large amount of unlabeled data during training. The computer will understand and develop the algorithm based on the labelled data provided and attempt to predict the labels for the new data.

• Unsupervised learning: Unsupervised machine learning techniques detect patterns from datasets without using known or labelled outcomes as input. Unsupervised learning, unlike supervised learning, cannot be used for a regression or classification problem because no information about the output data is available. As a result, unsupervised learning was employed to determine the data's underlying pattern.

The following are examples of unsupervised learning:

• Clustering: Clustering divides a set of data into subsets so that observations in the same cluster are of the same type.
• Clustering Algorithm with K Means: It is used to categorize unlabeled data, which is data that does not have established categories or groups. The method finds groups within the data, with the quantity K representing the number of groups found. Based on the attributes provided, it works iteratively to assign each data point to one of the K groups.

• Dimension reduction: Dimension reduction reduces the amount of variables that must be evaluated to find the necessary information.

• Reinforcement Learning: Reinforcement learning is a sort of dynamic programming that uses a reward and punishment system to train algorithms. A reinforcement-learning algorithm, also known as an agent, learns by interacting with its surroundings. When the agent performs successfully, they are rewarded, and when they perform wrong, they are penalized. As a result, it learns from previous experiences and begins to adjust its technique in reaction to the scenario to obtain the most significant possible results.
• Artificial neural networks (ANNs) are computing systems inspired by human biological brain networks. An ANN is built from a network of connected units or nodes known as artificial neurons, loosely modelled after the neurons in the human brain. Like synapses in a human brain, each link can send a signal to other neurons. ANNs also learn by example and experience. They are particularly effective for modelling non-linear connections in high-dimensional data or when the relationship between the input variables is difficult to interpret.