Images by the author | Illustrated characters
# introduction
From email spam filters to music recommendations, machine learning algorithms enhance everything. But they probably don't have to be complex black boxes. Each algorithm is essentially a different approach to finding patterns in your data and making predictions.
In this article, you will learn about important machine learning algorithms that all data experts should understand. For each algorithm, explain what it is doing and how it works in plain language, then explain what it should be used afterwards and when should not be used. Let's get started!
# 1. Linear regression
What is it: Linear regression is a simple and effective machine learning algorithm. To predict continuous values, find the best line for your data points.
How it works: Imagine trying to predict home prices based on area. Linear regression attempts to find the best line that minimizes the distance between all data points and lines. The algorithm uses mathematical optimization to find the optimal gradient and intercept for your data.
Where to use it:
- Forecast sales based on advertising spending
- Stock price estimate
- Predict demand
- Problems that expect a nearly linear relationship
If convenient: If your data has a clear linear tendency and you want interpretable results. It's also great if your data is limited or you need quick insights.
If not: Linear regression is not the best model if the data has complex, nonlinear patterns, or outliers and dependent features.
# 2. Logistic Regression
What is it: Logistic regression is also simple and is often used in classification problems. Predicts probability, values within range [0,1].
How it works: Instead of drawing a straight line, logistic regression uses an S-shaped curve (SIGMOID function) to map any input to a value between 0 and 1. This creates a probability score that can be used for binary classification (yes/no, not spam/spam).
Where to use it:
- Email Spam Detection
- Medical diagnosis (no illness/no illness)
- Marketing (Customers do not buy/buy)
- Credit Authorization System
If convenient: If estimates of probability are required along with predictions, then there is either a linearly separable data or a fast, interpretable classifier is required.
If not: For complex, nonlinear relationships, or when there are multiple classes that cannot be easily separated.
# 3. Decisions Tree
What is it: A decision tree works exactly like human decision-making. They ask a series of yes/no questions to arrive at a conclusion. Think of it as a flow chart for making predictions.
How it works: The algorithm starts with the entire dataset and finds the best questions to split into more uniform groups. Repeat this process to create branches (or stop based on defined criteria) until you reach a pure group. Therefore, the path from root to leaf is a decision rule.
Where to use it:
- Medical Diagnosis System
- Credit Scoring
- Function selection
- Domains that require naturally explained decisions
If convenient: If you want interpretable results, if you have mixed data types (numbers and categories), or if you want to understand which features are most important.
If not: They often tend to be overly unstable and unstable (small data changes can create very different trees).
# 4. Random Forest
What is it: If one decision tree is good, many trees are better. Random Forest combines multiple decision trees for more robust predictions.
How it works: Create multiple decision trees. Each decision tree is trained on a random subset of data using a random subset of features. For prediction, it receives votes from all trees and uses a majority victory for classification. Use the mean of the regression problem so that you can already guess.
Where to use it:
- Classification issues such as network intrusion detection
- E-commerce recommendations
- Complex Predictive Tasks
If convenient: If you need high precision without much tuning, if you need to handle missing values, or if you want a ranking of the importance of the feature.
If not: If you want very fast predictions, you need limited memory or highly interpretable results.
# 5. Supports vector machines
What is it: Support Vector Machines (SVMs) find the optimal boundaries between different classes by maximizing margins. The margin is the distance between the boundary and the nearest data point from each class.
How it works: Think of finding the best fence between two neighbours. SVMs don't just find fences. It finds the most possible from both neighborhoods. For complex data, “kernel tricks” are used to operate at a higher dimension that allows for linear separation.
Where to use it:
- Multi-class classification
- Small to medium data sets with clear boundaries
If convenient: If there is a clear margin between classes, limited data, or higher dimensional data (such as text). It is also memory efficient and versatile with a variety of kernel functions.
If not: If you want very large datasets (slow training), noisy data with overlapping classes, or estimates of probability.
# 6. K-means clustering
What is it: K-Means is an unsupervised algorithm that groups similar data points without knowing the “correct” answer in advance. By putting together similar items, it's like organizing a messy room.
How it works: Specifies the number of clusters (k), and the alignment of the algorithms k Randomly center of gravity in data space. Next, assign each data point to the nearest centroid and move the centroid to the center of the assigned point. This process is repeated until the center of gravity stops moving.
Where to use it:
- Customer segmentation
- Image quantization
- Data Compression
If convenient: If you need to discover hidden patterns, segment your customers or reduce the complexity of your data. It's simple, fast and works well with spherical clusters.
If not: If the clusters have different sizes, densities, or non-spherical shapes. It also does not robust to outliers and requires K to be specified in advance.
# 7. Naive Bays
What is it: Naive Bayes is a probabilistic classifier based on Bayes' theorem. It is called “naive” because it assumes that all functions are independent of each other.
How it works: The algorithm uses the Bayesian theorem to calculate the probability of each class given input functionality. This combines previous probability (the general one for each class) and likelihood (the possibility of each feature in each class) to make predictions. Despite its simplicity, it is extremely effective.
Where to use it:
- Send an email to spam filtering
- Text classification
- Emotional analysis
- Recommended systems
If convenient: If training data is limited, fast predictions, manipulation of text data, or a simple baseline model is required.
If not: If the feature independence assumption is severely violated, there are continuous numerical features (Gaussian naive Bayes may be useful), or the most accurate predictions are required.
# Conclusion
The algorithms discussed in this article form the basis of machine learning, including: Linear regression for continuous prediction. Logistic regression of binary classification. A decision tree for interpretable decisions. Random forest for robust accuracy. SVM for simple but effective classification. k-means for data clustering; naive Bayes for probabilistic classification.
Start with simpler algorithms, understand your data and use more complex methods if necessary. The best algorithms are often the simplest algorithms that solve problems effectively. Understanding when to use each model is more important than remembering technical details.
Bala Priya c I am an Indian developer and technical writer. She likes to work at the intersection of mathematics, programming, data science and content creation. Her areas of interest and expertise include Devops, Data Science and Natural Language Processing. She enjoys reading, writing, coding and coffee! Currently, she is committed to learning and sharing knowledge with the developer community by creating tutorials, how-to guides, opinions fragments and more. Bala also creates compelling resource overviews and coding tutorials.
