Photo by Author | Ideogram
. Introduction
From your email spam filter to music recommendations, the machine learning algorithm strengthens everything. But they are considered to be considered to be black boxes. Each algorithm is primarily a different approach to the search and predictions of samples in the data.
In this article, we will learn the necessary machine learning algorithms that should be understood by every data professional. Each algorithm’s i’ll, I will explain what it does and how it works in a simple language, then you should use it and when you should not. Let’s start!
. 1. The linear regression
What is it: Linear regression is an easy and efficient machine learning algorithm. It constantly looks for the best straight line through your data points to predict values.
How does it work: Imagine that you are trying to predict square footage -based home prices. Linear regression strives to find the best footline that reduces the distance between all your data points and line. The algorithm uses math correction to find slopes and breaks that best fit your data.
Where to use it:
- Predicting sales based on advertising costs
- Estimating stock prices
- Demand for forecasting
- Any problem where you almost expect a linear relationship
When it is useful: When your data has a clear linear trend and you need interpretation results. This is also great when you have limited data or need instant insights.
When it is not: If your data has complex, non -linear samples, or it has outline and dependent features, linear regression will not be the best model.
. 2. Logistics regression
What is it: Logistics regression is also easy and is often used in classification issues. It predicts possibilities, values in the limit (0,1).
How does it work: Instead of drawing a straight line, the logistics registration uses an S -shaped curved letter (sigmide function) to make any input map between 0 and 1. This produces a possibility that you can use for binary rating (yes/no, not spam/spam).
Where to use it:
- E -mail spam detection
- Medical diagnosis (illness/no disease)
- Marketing (will buy/buy/buy customer)
- Credit approval system
When it is useful: When you need an estimate of probability with your forecasts, obtain separate data, or sharp, interpretation rating.
When it is not: Complex, non -linear relationships or when you have multiple classes that are not easily separated.
. 3. The decision tree
What is it: Decision trees work just like human decision making. They ask a series of questions to come to a conclusion yes/no. Think about it as a flu chart that predicts.
How does it work: The algorithm starts with your entire dataset and finds the best question to divide it into more uniform groups. It repeats this process, the creation of branches until it reaches pure groups (or stops on the basis of default standards). Therefore, the routes from the roots to the leaves are the rules of the decision.
Where to use it:
- Medical diagnosis system
- Credit scoring
- The choice of the feature
- Any domain where you need naturally described decisions
When it is useful: When you need extreme translation results, keep mixed data types (numerical and category), or want to understand which features are the most important.
When it is not: They often suffer from being more appropriate, unstable (changes in small data can produce very different trees).
. 4. The random forest
What is it: If a decision tree is good, many trees are better. Random Forest has added several decision trees to make stronger predictions.
How does it work: It makes several decisive trees. Each tree of judgment is trained on a random sub -set of data using random sub -sets of features. Predicts IT, it votes from all the trees and uses the majority victory for rating. As you can already guess, it uses average in reactionary issues.
Where to use it:
- Classification issues such as detecting interference in a network
- E -commerce recommendations
- Any complicated work of prediction
When it is useful: When you want high accuracy without too much tuning, the lost values need to be handled, or the importance of the feature.
When it is not: When you need very fast predictions, the memory is limited, or the most explanatory results are needed.
. 5. Support vector machines
What is it: Support vector machines (SVM) find a maximum limit between different classes by maximizing margin. The margin range and the closest data points to each class is the distance.
How does it work: Think about it to find the best fence between the two palaces. SVM does not just find any fence. It has been found a person who is from both palaces as far as possible. IT of complex data, it uses “kernel tricks” to work in high dimensions where linear separation is possible.
Where to use it:
- Multi -class rating
- On small to medium datasis with clear limits
When it is useful: When you have a clear margin between classes, limited data, or high -dimensional data (such as text). It is also memory efficient and versatile with various kernel functions.
When it is not: With huge datases (slow training), noise data with overlaping classes, or when you need an estimate of possibility.
. 6. K-Means clustering
What is it: K-Means is an undisclosed algorithm that collects similar data points without knowing the “right” answer. This is equivalent to organizing a dirty room by placing similar items together.
How does it work: You specify the number of clusters (k), And places of algorithm k Centerodes in place of your data space. It then assigns each data point to the nearest Center and transmits the center to the center of its assigned points. This process repeats until the center does not move.
Where to use it:
- Customer’s distribution
- Image quantization
- Data compression
When it is useful: When you need to discover hidden patterns, class users, or reduce the complexity of the data. It works well with simple, fast, and globalular clusters.
When it is not: When clusters contain different sizes, density, or non -chronic shapes. It is not strong for outsiders and you need to explain before K.
. 7. Bid Twenty -two
What is it: Bid is a possibility based on the theory of twenty -two. This is called “bid” because it assumes that all features are free from each other, which is rarely true in real life but practically works amazingly.
How does it work: The algorithm calculates the possibilities of each class, in which the features of the input are used by using the theory of Bais. It combines the possibility of predictions (how much common in each class) (how much is common in each class) (how much the possibility of each feature is common for each feature). Despite its simplicity, it is noteworthy.
Where to use it:
- E -mail spam filtering
- The text rating
- Analysis of emotions
- Recommendation system
When it is useful: When you have limited training data, need sharp predictions, work with text data, or want a simple baseline model.
When it is not: When a severe violation of the characteristic assumption of the characteristic, you have constant numerical features (though the Gosi bid can help base), or potentially require extremely accurate predictions.
. Conclusion
The algorithm we have discussed in this article form the basis of machine learning, including: linear regression for continuous predictions. Logistics Regression for binary rating; Decision tree for explanatory decisions; Random forests for strong accuracy; SVM for simple but efficient rating; K-Means for data clustering; And bid for potential rating.
Start with easy algorithm to understand your data, then use more complicated ways when needed. The best algorithm is often the easiest that solves your problem effectively. It is more important than memorizing technical details to use each model.
Pray Ca Is a developer and technical author from India. She likes to work at the intersection of mathematics, programming, data science, and content creation. The fields of interest and expertise include dupas, data science, and natural language processing. She enjoys reading, writing, coding and coffee! Currently, they are working with the developer community to learn and share their knowledge with the developer community by writing a lesson, how to guide, feed and more. The above resources review and coding also engages lessons.