Basic Terminology in Machine Learning

1. Training

Meaning: Training is the phase where we teach the model by showing it a dataset with inputs and outputs (called training data).
The algorithm adjusts its internal values (parameters) to reduce errors.
Analogy: A student learning math by practicing solved examples.

Figure idea:
Imagine a scatter plot of house sizes vs. house prices. During training, the algorithm tries to draw the best-fitting line/curve through those points.

2. Validation

Meaning: After training, we need to check how well the model works on data it has not seen. This is done using a validation dataset.
Helps in:
- Choosing the best model
- Tuning hyperparameters (like learning rate, tree depth, number of neighbors in KNN).
Analogy: A student taking a practice test before the final exam to identify weaknesses.

Figure idea:
Training set → used to fit the curve
Validation set → checks if the curve is correct or needs adjustment.

3. Testing

Meaning: Testing is the final stage where the model is evaluated on completely unseen data (test set).
Purpose: To measure how well the model will perform in the real world.
Analogy: The final exam after preparation — no questions are repeated from practice.

Figure idea:
A flow:
Data → (Split into) Training (70%) → Validation (15%) → Testing (15%)

4. Model

Meaning: A model is the result of training an algorithm. It represents the relationship between inputs (features) and outputs (labels).
Example: A Linear Regression model might learn the relationship:
$\text{Price} = m \cdot (\text{Size}) + c$
Analogy: The trained brain of a student who has learned patterns from examples.

Figure idea:
Input (House Size) → [Model] → Output (Predicted Price)

5. Parameters

Meaning: Parameters are the internal values that the model learns from training data.
They define the model’s prediction behavior.
Examples:
- In Linear Regression → slope (m) and intercept (c)
- In Neural Networks → weights and biases
Analogy: A student’s memory of formulas after studying.

Figure idea:
Neural network diagram with arrows (connections) having weights labeled — these are parameters.

6. Overfitting

Meaning: When a model learns the training data too well, including noise and irrelevant details.
It performs great on training data but poorly on new data.
Analogy: A student memorizing answers without understanding concepts → does well in practice, fails in the real exam.

Figure idea:

Training data points with a very zig-zag curve passing through every point → shows overfitting.

7. Underfitting

Meaning: When a model is too simple to capture patterns in data.
It performs poorly on both training and test data.
Analogy: A student who did not study enough → fails in both practice and final exams.

Figure idea:

Training data points forming a curve, but the model tries to fit a straight line → underfitting.

8. Real-life Applications of ML

Machine Learning is everywhere. Some examples:

Healthcare → Predicting diseases, analyzing medical scans (X-rays, MRIs).
Finance → Fraud detection, stock price prediction, loan approvals.
Education → Personalized learning apps, automatic essay grading.
E-commerce → Recommendation systems (Amazon suggesting products).
Transportation → Self-driving cars, traffic predictions, smart maps.
Social Media → Detecting fake accounts, personalized news feeds.
Daily Life → Voice assistants (Siri, Alexa), face recognition on phones.

Summary :-

Training → Teaching model with known data
Validation → Fine-tuning model with new but not final data
Testing → Final performance check on unseen data
Model → Learned mathematical representation
Parameters → Internal values learned during training
Overfitting → Too complex → memorizes data
Underfitting → Too simple → misses patterns
Applications → Used in healthcare, finance, transport, education, daily life