The most commonly used machine learning algorithms are:

**Logistic Regression** This classification estimates discrete values (binaries like 0-1, true-false, or yes-no) based on independent variables. It predicts an events occurrence probability by putting data in logit function because of which it’s known as a logit regression. The odds of the outcomes are modeled as predictor variables’ linear combination.

**Linear Regression** Used to estimate values (like the cost of houses, total sales, number of calls etc) based on a continuous variable, Linear Regression establishes a relationship between independent and dependent variables by fitting the best line.

**Simple Linear Regression** is the regression based on a single variable while multiple linear regression is based on more than one variable.

**Naive Bayes** This is a technique for classification that uses the Bayes theorem assuming independence of predictors. The Naive Bayes assumes that a specific feature’s presence in class is not related to another one’s presence. It provides a method to calculate posterior probability.

**Decision Tree** It belongs to a supervised algorithm that experts use mostly for classification issues. It works in categorical as well as continuous related variables. This algorithm works by splitting the population into sets that are homogeneous. This functions on the basis of attributes of most significance/free variables to form unique groups. For dividing the population into heterogeneous groups, different techniques like Information **Gain, entropy, Gini,** and** Chi-square** are used.

**SVM (Support Vector Machine)** This algorithm works on the basis of plotting each item of data as point space of n number of dimensions (where n represents features number) and the value of the feature becomes a specific coordinate value.

**K-Means** This unsupervised algorithm is designed to solve the clustering problem. It follows an easy way to classify a data set into a certain number of clusters. The data points are homogeneous inside a cluster and to peer groups heterogeneous.

**Random Forest** This is the term used to mark a group of decision trees. It works on the basis of a collection of decision trees.

**Dimensionality Reduction Algorithms** There exponential increase in data capturing at all stages in the past five years has resulted in the application of several algorithms.

**Gradient boosting algorithms** Several algorithms are classified into this group including XGBoost, GBM, CatBoost, and LightGBM. Some of these are boosting algorithms that deal with big data for high power predictions. Others are tree-based algorithms designed for distributed use and efficiency that allows faster training, lower memory use, higher accuracy, and handling of huge amounts of data. Boosting algorithms play a key role in dealing with bias-variance trade-off. Bagging algorithms on the other hand only control for a model’s high variance and boost controls of both aspects like variance and bias. It is considered more effective.

**Supervised learning** is the training of a program with a data set of training examples with related correct labels. The algorithm proceeds to learn the link between the images and the relevant numbers and apply that relationship so learned to classify images that are fresh and are without labels.

**Regression and classification** Regression predicts a target variable in numerical terms like the sale price of a property On the other hand, in the case of classification a label is assigned to a pattern and based on that label similar patterns are classified under a label and identified.

**Reinforcement learning **Machine learning crosses path with human behaviorist psychology in reinforcement learning. It’s goal-oriented and based on the interaction of the algorithm with the environment. Reinforcement learning is producing a variety of learning algorithms that have a use in different applications. It is the act of learning how to react to situations.

**Confusion matrix,** sometimes called error matrix, is used in statistical classification. It’s the table that describes the model of classification, the classifier – for a test data set with known correct values. It makes possible the visualization of an algorithm’s performance. The confusion matrix facilitates plotting the confusion among classes.

**The gradient descent** **approach** is adopted for learning the parameters to upgrade the quality of supervised learning. Gradient descent’s goal is to progressively proceed towards minimum loss function by getting better by iterative approximation. Use of more training data is one common way of curing overfitting. The other regularization.