# Machine Learning Algorithms Every Beginner Should Know

In a world where almost every task is getting automated, manual tasks are slowly changing. Today, machine learning algorithms are allowing computers to perform surgeries, play chess and get more personal. These algorithms can even predict match outcomes, with sites like https://www.oddsshark.com/nhl/computer-picks offering computer picks on different games.

The main feature of this revolution is how computing techniques and tools have become democratized. In the last half a decade, data scientists have developed sophisticated data-crunching algorithms using advanced techniques and the results have been incredible. Here are popular machine learning algorithms for beginners.

Popular Machine Learning Algorithms

There are three common types of machine learning algorithms; supervised learning algorithms, unsupervised learning algorithms, and reinforced learning. Supervised learning algorithms use training data to understand the mapping functions and turn the input (X) variables to the output (Y) variable – Y=f(X). That allows you to accurately generate the outputs when new inputs are given.

Unsupervised learning algorithms are used when only the input variables are given without a corresponding output variable. These models use unlabeled training data to develop the underlying data structure.

Reinforcement algorithms, on the other hand, allow an agent to choose the next best action based on the current state of learning behaviors to maximize rewards. These models learn optimal options through trial and error. For instance, a reinforced algorithm used to play a game will start with a random mover and learn where the in-game character needed to move to maximize the total points.

Top Five Machine Learning Algorithms For Beginners

Linear Regression

In machine learning, there are a set of input variables that determine the output variables. With linear regression algorithms, that relationship between the input and output variables is expressed using the equation Y= a + bX. That means the goal of this model is to find the values of the coefficients a and b, where Y is the output and X represents the input.

Logistic Regression

While linear regression predicts continuous values, logistic regression offers discrete values after using a transformational function. That means logistic regression works best for binary classifications (data sets where y=1 or 0, with 1 being the default class. For instance, when trying to predict whether an event will happen or not, only two possibilities are there so you use 1 to denote that it occurs and 0 denotes that it won’t.

This algorithm derives its name from the transformation function it uses, also called the logistic function – h(X) = 1/(1+eX). The output is in the form of the probabilities of its default class, which explains why the output ranges from 0-1. Its goal is to use the training data to get the coefficients of b1 and b0 to minimize the error (distance) between its prediction and the actual result. The coefficients are derived using the Maximum Likelihood Estimation technique.

Apriori Algorithms

The Apriori algorithm is most useful with a transactional database for mining frequent itemsets and generating association rules. The algorithms are popular in market basket analysis when checking for combinations of products frequently co-occurring in a database. Generally, we use the association rule that if a customer buys item X, then he buys item Y as X->Y.

For example, if a customer buys sugar and milk, he’s likely to buy coffee. That can be written as (Sugar, Milk) -> coffee and association rules are created after crossing the confidence and support threshold. That can be written as;

Rule: X-> Y

Support = frq(X,Y)/N

Confidence = frq(X,Y)/frq(X)

Lift = Support/sup(X) x Supp(Y)

The support helps to cut the number of candidate sets considered in the frequent itemset generation. The Apriori principle guides the support measure. This principle states that in a frequent itemset, all the subsets should also be frequent.

Principal Component Analysis (PCA)

PCA is used when making data easy to visualize and explore by reducing the variables. That can be done by putting the maximum variance in your data in a new coordinate system that uses axes known as “principal components”.

Every component in the model is a linear combination of your original variables and they’re orthogonal to each other. The orthogonality between these components shows that the correlation between such components is zero.

The first component determines the direction of your maximum variability in your data, while the second captures the remaining variance. However, there are uncorrelated variables in the first component. All the successive principal components capture the remaining variance and remain uncorrelated with the previous components.

Bagging with Random Forests

The first step with this algorithm is creating several models with data sets developed using the Bootstrap Sampling technique. With this technique, every generated training set comprises random subsamples from your original data sets. These training sets have the same size as your original data sets, although some records are repeated several times some don’t appear at all. Nonetheless, the whole original data set can be used as a test set.

The second step involves creating several models using the same algorithm, but with different training sets. That’s where Random Forests are involved, where you randomly choose a selection of features for developing the best split. The reason behind this randomness is that when the decision tree chooses the best feature, it results in correlated predictions and a similar structure.