🎯 Welcome to Machine Learning Fundamentals. This course assumes you're already comfortable with core Python, NumPy arrays, and pandas DataFrames — if any of that feels shaky, BitWithBite's Data Science with Python course has a full treatment, and we'll link back to it occasionally for a quick syntax recap. From here on, we go deep on ML itself: what it is, the major families of ML problems, the standard workflow every model follows, the preprocessing steps that make or break a model, and the single most important failure mode — overfitting — before moving into real algorithms starting with linear regression in Section 2.
Section 1
What Is Machine Learning?
Machine learning is a way of building software that improves its behavior by learning patterns from data, instead of following rules a human wrote by hand. In traditional software, a developer writes explicit logic — "if the transaction is over $10,000 and from a new device, flag it." In machine learning, you instead give the system a large number of examples, and an algorithm searches for the statistical patterns that separate one outcome from another.
The Core Shift
Rules + Data → Answers ⟷ Data + Answers → Rules
Traditional code executes rules a human wrote. A trained ML model IS the rules — discovered automatically from labeled or unlabeled examples, then reused to make predictions on data it has never seen.
The output of this process is called a model: a set of learned numeric parameters that, when combined with new input data, produces a prediction. Training a model means running an optimization procedure that searches for the parameter values that make the model's predictions match the known outcomes as closely as possible, across the whole training dataset at once — not just memorizing individual rows.
📝
"Learning" means optimization, not comprehension
A model doesn't understand what a "fraudulent transaction" or a "cat" is the way a person does. It finds numeric parameters that minimize the gap between its predictions and the correct answers on the training data. That's a mathematical optimization problem — a much larger-scale version of finding the slope and intercept of a line of best fit — not a form of reasoning.
🔁
Quick recap, if you need it
Everything in this course leans on NumPy arrays (X as a 2D array of features, y as a 1D array of targets) and pandas DataFrames for loading and inspecting tabular data. If df.head(), boolean indexing, or NumPy array shapes aren't familiar yet, BitWithBite's Data Science with Python course (Sections 1–2) covers them from scratch. We won't re-teach that syntax here.
Section 2
The Three Types of Machine Learning
Nearly every ML problem falls into one of three broad categories, defined by what kind of data and feedback the algorithm has access to during training.
🏷️
Supervised Learning
Every training example is paired with the correct answer. The model learns to map inputs to known outputs.
🔍
Unsupervised Learning
No correct answers are provided. The model finds structure, groups, or reduced representations on its own.
🎮
Reinforcement Learning
An agent takes actions in an environment and learns from reward or penalty signals over time.
Supervised Learning
In supervised learning you have a dataset where each row has input features (X) and a known target value (y) — a labeled dataset. The model's job is to learn the mapping from X to y well enough to predict y for new rows it hasn't seen. Supervised learning itself splits into two sub-tasks depending on what kind of target you're predicting:
📈
Regression
The target is a continuous number — a house price, a temperature, a stock's next-day return. Covered starting in Section 2 of this course with linear regression.
🏷️
Classification
The target is one of a fixed set of categories — spam/not-spam, or which of several species an image shows. Covered in Section 3 with logistic regression and beyond.
Unsupervised Learning
Here you only have X — no labeled target at all. The model looks for structure that already exists in the data without being told what to look for. Two common unsupervised tasks:
🧩
Clustering
Groups similar rows together — for example, segmenting customers into behaviorally similar groups with no pre-defined labels. Algorithms like K-Means fall here.
📉
Dimensionality Reduction
Compresses a large number of features down to a smaller set that still captures most of the important variation — useful for visualization and for speeding up other models. PCA is the classic example.
Reinforcement Learning
Reinforcement learning is a different setup entirely — there's no fixed dataset of examples up front. Instead, an agent interacts with an environment: it takes an action, observes the result, and receives a reward or penalty. Over many attempts, it learns a strategy (a "policy") that tends to maximize cumulative reward. This is the family of techniques behind things like game-playing agents and robotic control. This course focuses on supervised and unsupervised learning — reinforcement learning is a large enough topic that it's worth knowing the name and the shape of the idea, but it isn't covered hands-on here.
⚠️
The labels are the whole difference
If your dataset has a clear column you're trying to predict — a price, a category, a yes/no — that's supervised learning. If there's no such column and you're looking for structure within the data itself, that's unsupervised learning. If there's no dataset at all, just an agent acting inside an environment over time, that's reinforcement learning.
Section 3
Real-World Application Categories
Machine learning shows up across a small number of recurring problem shapes, each of which spans many industries. Here are four broad categories worth knowing by name — described generically, since the exact techniques and performance depend heavily on the data and problem at hand.
🎬
Recommendation Systems
Predict which items a user is likely to want next, based on their past behavior and the behavior of similar users. Underlies "you might also like" style features across retail, streaming, and content platforms.
Supervised + Unsupervised
🛡️
Fraud & Anomaly Detection
Flag transactions, logins, or events that look statistically unusual compared to normal behavior. Often framed as a classification problem, sometimes as unsupervised anomaly detection when fraud examples are rare or unlabeled.
Classification
👁️
Computer Vision
Extract information from images or video — identifying objects, reading handwritten digits, detecting defects on a production line. Built on models that treat pixel values as input features.
Classification / Regression
💬
Natural Language Processing (NLP)
Work with text — classifying sentiment, detecting spam, summarizing documents, powering search and chatbots. Text is converted into numeric features before any of the algorithms in this course can use it.
Classification / Sequence Models
✨
Same toolbox, different framing
Notice that fraud detection and spam filtering are both, underneath, binary classification problems. Recommendation and clustering both lean on finding similarity between rows. Once you understand the core algorithm families this course covers, you'll recognize the same shapes recurring across very different-sounding applications.
Section 4
Lesson Summary
✅Machine learning is software that learns patterns from data instead of following explicit hand-written rules.
✅The three families are supervised (labeled data), unsupervised (no labels), and reinforcement learning (reward-driven agents).
✅Supervised learning splits into regression (predicting numbers) and classification (predicting categories).
✅Unsupervised learning covers clustering and dimensionality reduction.
✅Real-world applications cluster into recurring shapes: recommendation systems, fraud detection, computer vision, and NLP.
🧩 Knowledge Check — Lesson 1
4 questions on the core ML vocabulary before you move on.
1. A dataset of customer transactions has NO "fraud yes/no" column, and the model groups transactions into clusters on its own. What type of learning is this?
2. Predicting the exact resale price of a used car is an example of which supervised sub-task?
3. Which technique is used to compress a large number of features into a smaller set while retaining most of the important variation?
4. An agent playing a game learns a strategy purely from win/loss reward signals, with no labeled dataset at all. What type of learning is this?
💪
Try It Yourself — Lesson 1
Get comfortable with the vocabulary · Intermediate Level
No models to train yet — this lesson's task list locks in the categories before Section 2 puts them to use.
Task 1: Classify five real-world problems 🏷️
For each of these, decide whether it's supervised (and if so, regression or classification), unsupervised, or reinforcement learning: (a) predicting tomorrow's electricity demand from historical usage, (b) grouping news articles into topics with no pre-defined categories, (c) predicting whether a tumor is malignant or benign from scan measurements, (d) a robot learning to walk through trial and error, (e) compressing a dataset of 200 features down to 10 for visualization.
Task 2: Map an application to a category 🔍
Pick one real-world ML application you use regularly (a streaming recommendation, a spam filter, a photo app that tags faces) and write two or three sentences on which of the four application categories from Section 3 it fits, and whether the underlying task is most likely classification, regression, or clustering.
Task 3: Explain the core shift in your own words 📝
In two or three sentences, explain the difference between "Rules + Data → Answers" and "Data + Answers → Rules" from Section 1, using an example that wasn't used in this lesson.
Task 2: Most consumer-facing recommendation and personalization features lean on a mix of classification (will the user click this?) and similarity/clustering (which users or items look alike?).
Task 3: A good answer names a task where a human would traditionally hand-write rules, and explains what would instead be "learned" if it were treated as an ML problem — e.g. deciding loan approvals from historical repayment records instead of a fixed checklist.
Finished this lesson?
Mark it complete to track your progress.
🎉
Lesson 1 Complete!
You now know the three types of ML, the classification/regression/clustering split, and the application categories ML shows up in. Next: the standard end-to-end ML pipeline every project follows.
Module 01 of 24
Section 1 — What is Machine Learning?