Introduction to Machine Learning Concepts: Data, Features, Labels, Training, and Common Algorithms

Machine learning often feels lofty because of its terminology; this article explains the core concepts in plain language.

1. Data

In programming we frequently use database; a row in a database corresponds to a record, which contains many attributes. These attributes are directly related to machine learning features.

2. Features

Each attribute of a record is called a feature in machine learning. Features are divided into discrete and continuous types.

2.1 Discrete Features

Examples from high‑school math: continuous vs discrete. In a sample Bird class we define attributes such as weight, length, fins, color, and type. These attributes are also called features.

Color is an enumerated feature.

Weight and length are numerical features.

Fins are boolean features (also called binary features).

Type is the target variable.

2.2 Continuous Features

Continuous data are typically interval data, e.g., temperature ranges like 19~28°C or water freezing point -4~−∞°C.

3. Dimensionality

The number of features equals the dimensionality. For the Bird class with four features, each data vector has four dimensions.

4. Target Variable (Label)

The target variable is the value the algorithm predicts, e.g., y = x + 1. In classification it is usually categorical, while in regression it is continuous. The target must be present in the training set to relate features to the label.

5. Training

Training (algorithm training) feeds a large set of labeled data to produce an algorithm model. The data used for training is called the training set.

Example: feeding data to an algorithm yields f(x) = x/2, which becomes the model.

6. Datasets

6.1 Training Dataset

The training dataset provides input data for model training.

6.2 Test Dataset

The test dataset is used after training to evaluate the model’s performance; it must contain the target variable to measure accuracy and must not be used during training.

7. Normalization

Normalization (numeric scaling) rescales features to the 0~1 range to avoid dominance of large‑scale attributes in distance calculations.

Simple formula: new_value = (old_value - min) / (max - min) where min and max are the feature’s minimum and maximum in the dataset.

8. Supervised Learning

Supervised learning uses labeled data. It requires knowledge of both features and target variable.

8.1 Classification

Classification assigns data to predefined categories; it is a primary task of machine learning.

Banana  Orange  Fork  Cola
|       |       |     |
Fruit   Fruit   Utensil  Drink

8.2 Regression

Regression predicts continuous values, e.g., fitting a curve to data points.

8.3 Common Supervised Algorithms

(Illustration omitted.)

9. Unsupervised Learning

Unsupervised learning works with only features (no target). It discovers structure such as clusters.

9.1 Clustering

Clustering groups similar data points into clusters.

Apple  Orange  Spoon  Fork  Cola  Sprite
 \      /      \    /    /      \
  Fruit   Utensil   Drink

9.2 Density Estimation

Estimating the statistical distribution of data.

9.3 Common Unsupervised Algorithms

(Illustration omitted.)

10. Choosing the Right Algorithm

Select an algorithm based on the task and data type: use supervised learning for predicting a target; if the target is discrete, choose a classification algorithm; if continuous, choose regression. If no target is needed, consider unsupervised methods such as clustering or density estimation.

11. Machine Learning Development Steps

Collect data (web crawling, sensors, public datasets).

Prepare input data (formatting, discretization).

Analyze data (missing values, outliers, visualizations).

Train algorithm (choose appropriate model; unsupervised methods skip training).

Test algorithm (evaluate model performance).

Deploy algorithm (engineer the model for production).

12. Recommended Language and Libraries

Python is recommended for its powerful libraries and simplicity.

Common libraries: numpy for vector/matrix operations and matplotlib for 2D/3D plotting.

13. Numpy Common Operations

13.1 Create Random Data

import numpy
numpy.random.ramd(4,4)

13.2 Convert Array to Matrix

import numpy
random_array = numpy.random.ramd(4,4)
random_matrix = numpy.mat(random_array)

13.3 Matrix Inverse

import numpy
random_matrix = numpy.mat(numpy.random.ramd(4,4))
random_matrix.I

13.4 Matrix Multiplication

import numpy
random_matrix = numpy.mat(numpy.random.ramd(4,4))
result = random_matrix * random_matrix.I

13.5 Create Identity Matrix

import numpy
matrix = numpy.eye(4)

- END -