Full course · free on YouTube · 15 h 28 min

Machine Learning Fundamentals: free, 15.5 hours

Two recorded live class sessions, 8 hours and 7.5 hours, that start from probability and statistics and move through supervised learning, unsupervised learning, PCA and a first look at neural networks. Each session mixes slide explanations, hand calculations and Python code, and sets homework and challenges.

35 chapters · 2 parts · published 20 Sept 2024 · 83,422 views as of 18 Nov 2026 · by Skill-Lync

Free. We ask for your name and phone once before it plays here · or watch on YouTube ↗

Who is this course for?

Mechanical and other engineering students or graduates who want to learn machine learning with Python, with basic maths and statistics covered inside the course.

What will you learn?

  • Calculate probabilities, variance, standard deviation and Z-scores, and explain why standardising puts different variables on one scale
  • Fit simple and multi-linear regression, measure R-square, and write gradient descent in Python with mean square and mean absolute error
  • Build logistic regression, decision trees, random forests, SVM and KNN models in scikit-learn and compare them
  • Read a confusion matrix, precision, recall, F1 score and the ROC curve, and spot overfitting and underfitting from training and test error
  • Cluster data with K-means and the elbow curve and with hierarchical clustering and dendrograms
  • Reduce features with PCA on the Iris dataset and understand how a neural network builds on logistic regression

What does each part of the course cover?

12 sections, in watching order. Click a timestamp to jump the player there. Many of the description's chapter titles do not match what the class covers (several are just 'Introduction to AI and ML using Python'), so the titles here are written from the video itself.

Part 1 · 8 h 3 min

  1. 4:021 h 44 min · 4 chapters

    Why Python, probability, sampling and Python loops

    Opens with why Python and why data matters, then works through coin-toss probability (two coins, three coins, 4C2/2^4, 100C50), population versus sample, descriptive and inferential statistics, mean, variance and standard deviation, random variables and Python for loops with range and indentation.

    Chapters
    • 4:02Course orientation, why Python, and probability with coin tosses
    • 37:29Sampling, descriptive and inferential statistics, mean, variance and standard deviation
    • 1:07:06Random variables, discrete and continuous, and Python basics
    • 1:27:23Python for loops, range and sample size; probability distributions
  2. 1:48:1652 min · 2 chapters

    Normal distribution, Z-scores and descriptive analytics

    Shows why normal curves for pressure, temperature and speed cannot be compared directly, then standardises x and y columns in Excel to mean 0 and standard deviation 1. Covers Z-tables and the 68% area, mean, median and mode, the four moments including skewness and kurtosis, and the four types of analytics.

    Chapters
    • 1:48:16Normal distribution and Z-scores (standard normal)
    • 2:18:38Four moments, the four types of analytics and AI versus ML versus DL
  3. 2:39:491 h 32 min · 4 chapters

    AI, ML and deep learning, linear regression and gradient descent

    Separates AI, machine learning and deep learning, supervised and unsupervised learning, and regression from classification. Derives the best-fit line from squared error, covers R-square and the learning rate in gradient descent, then moves to training and test splits, NumPy, pandas, multi-linear regression and coding gradient descent with for loops, mean square and mean absolute error.

    Chapters
    • 2:39:49Z-score probability problems, supervised vs unsupervised learning, regression vs classification
    • 2:57:11Linear regression: best-fit line, error, R-square and gradient descent
    • 3:17:01Train/test split, NumPy, pandas and multi-linear regression in Python
    • 3:42:52Coding gradient descent (MSE and MAE) and introducing logistic regression
  4. 4:12:091 h 26 min · 3 chapters

    Logistic regression, confusion matrix and overfitting

    Builds logistic regression from the linear equation passed through a sigmoid, with a 0.5 probability threshold, then explains the confusion matrix (true and false positives and negatives), human error versus training and test error, and generalisation. Ends with train_test_split in scikit-learn, log loss and label versus one-hot encoding.

    Chapters
    • 4:12:09Logistic regression, sigmoid, confusion matrix and error types
    • 4:42:39Generalisation vs overfitting, train_test_split and logistic regression in scikit-learn
    • 5:07:25Log loss, label encoding and one-hot encoding, multicollinearity, intro to decision trees
  5. 5:38:001 h 4 min · 2 chapters

    Decision trees, entropy and information gain

    Introduces decision trees with a fit or unfit example and a buying-a-computer table, explaining root, intermediate and leaf nodes and pure leaves. Calculates entropy and information gain by hand to pick the root node, then codes a DecisionTreeClassifier with a confusion matrix.

    Chapters
    • 5:38:00Decision trees: root node, leaf nodes, entropy and the buying-a-computer example
    • 6:10:51Information gain calculation and when to stop a tree
  6. 6:41:391 h 21 min · 3 chapters

    Random forest, evaluation metrics, over/underfitting and ROC

    Explains random forest as many trees on random samples with a majority vote, plus bootstrapping and the number of trees as a hyperparameter. Covers accuracy, recall, precision and F1 using a disease example, high bias versus high variance from training and test error, the 0.5 threshold, and the ROC curve with area under the curve.

    Chapters
    • 6:41:39Random forest, bootstrapping, majority rule, recall, precision and F1 score
    • 7:11:48Human error, training error and test error: spotting overfitting and underfitting
    • 7:37:21Why start with linear regression, the 0.5 threshold and an introduction to ROC curves

Part 2 · 7 h 25 min

  1. 0:0447 min · 2 chapters

    Support vector machines and the Iris dataset

    Finishes the ROC discussion, then explains support vector machines: separating lines, the optimal hyperplane, support vectors, the kernel trick and hyperparameters. Runs scikit-learn code on the Iris dataset (four features, three classes), trains an SVM and reads the result, 29 of 30 test records correct, in a confusion matrix.

    Chapters
    • 0:04ROC follow-up and support vector machines: hyperplane, support vectors, kernels
    • 19:51SVM on the Iris dataset in Python, with train/test split and confusion matrix
  2. 47:271 h 26 min · 3 chapters

    SVM parameters, K-nearest neighbours and comparing classifiers

    Discusses the C and gamma parameters of SVM, then K-nearest neighbours: distance to neighbours, the choice of K, noisy samples, majority vote and why KNN is called lazy learning. Codes KNeighborsClassifier with K=3, compares logistic regression, random forest, SVM and KNN on one dataset, and introduces k-fold cross validation.

    Chapters
    • 47:27SVM parameters C and gamma, kurtosis, and K-nearest neighbours theory
    • 1:19:15KNN distance calculation, effect of K, dummy variables and Z-score or min-max scaling
    • 1:46:37KNN in scikit-learn, comparing classifiers and k-fold cross validation
  3. 2:13:461 h 13 min · 3 chapters

    Kurtosis, unsupervised learning and K-means clustering

    Explains kurtosis with height and salary distributions, then unsupervised learning and examples of clustering and association rules. Covers the K-means steps (choose K, centroids, reassign, converge), the elbow curve for the optimal K, using cluster labels as classes for supervised learning, and why features need min-max scaling.

    Chapters
    • 2:13:46Kurtosis, finding a maximum in Python, and introduction to unsupervised learning
    • 2:32:29Clustering vs association rules, K-means steps and the elbow curve
    • 2:56:23Using K-means in scikit-learn, cluster labels as classes and min-max scaling
  4. 3:26:371 h 21 min · 3 chapters

    K-fold cross validation, hierarchical clustering and stratified splits

    Goes through k-fold cross validation, then hierarchical clustering: dendrograms, cutting branches to get clusters, bottom-up versus top-down, and single, complete and average linkage. Adds stratified splitting for imbalanced data, the final project brief, and a Python Fibonacci exercise with a for loop, break and input() type conversion.

    Chapters
    • 3:26:37K-fold cross validation and the start of hierarchical clustering with dendrograms
    • 3:56:50Hierarchical clustering: top-down vs bottom-up, linkage methods and the final project
    • 4:27:55Stratified splitting for imbalanced data and a Python Fibonacci and input() exercise
  5. 4:47:521 h 42 min · 4 chapters

    Principal component analysis on the Iris dataset

    Explains why to reduce features, then PCA: new features PC1 to PC4 ordered by variance, confidence thresholds, and the linear-algebra background. Applies it to the four Iris features after scaling, plots PC1 against PC2 (about 95% of variance) and compares logistic regression accuracy, 93.3% on the original data, with PCA components.

    Chapters
    • 4:47:52Final project brief and why we reduce features (PCA introduction)
    • 5:11:29PCA: principal components, variance retained and confidence
    • 5:36:58PCA on the Iris dataset in scikit-learn with standardisation
    • 6:09:12PCA results on Iris: explained variance and accuracy with and without PCA
  6. 6:30:1255 min · 2 chapters

    Neural networks: from logistic regression to hidden layers

    Builds a neural network from logistic regression: inputs, weights, a sigmoid activation per neuron, hidden layers and the zeroth, first and second layer notation. Revisits gradient descent weight updates, feed-forward versus recurrent networks, deep learning, training, dev and test sets, and bias versus variance.

    Chapters
    • 6:30:12Neural networks: neurons, weights, sigmoid activation and hidden layers
    • 6:51:49Gradient descent for neural networks, deep learning and bias versus variance

Is the Machine Learning Fundamentals In Depth course really free?

Yes. All 15 h 28 min of it is free on Skill-Lync's YouTube channel, with no payment, in 2 parts. To watch it on this page with the chapter guide, leave your name and phone number once; you can also watch it directly on YouTube. It was published on 20 Sept 2024 and had 83,422 views as of 18 Nov 2026. This page adds the chapter guide, a study plan and what to learn next.

How long does the Machine Learning Fundamentals In Depth course take to finish?

The video runs 15 h 28 min across 35 chapters. At one hour a day that is about 16 days; plan roughly twice that if you practise each exercise as you go. The longest section is "Why Python, probability, sampling and Python loops" (1 h 44 min), so give it its own days.

Does the free Machine learning course come with a certificate?

The video description has a "claim your certificate" link, which now opens this page. To ask about a certificate for this course, leave your name and phone number in the form below and a Skill-Lync counsellor will call you back. The paid Skill-Lync courses linked here include a certificate along with mentor-reviewed projects.

What should I learn after this course?

The next step is Machine Learning Fundamentals In Depth (₹40,000), then Post Graduate Program in Data Science and Machine Learning (₹2,75,000). The free course teaches the concepts; the paid Skill-Lync courses add graded projects, mentor feedback and, in the PG programs, placement support.

Do I need to know Python or machine learning before this course?

No prior machine learning is assumed. The instructor says the maths and statistics needed are very basic and will be covered in the course, and that every class includes hands-on Python. Coding starts from the basics, such as for loops and range, and homework challenges are set after each class.

Which Python libraries does the machine learning course use?

The code uses NumPy for numerical work, pandas data frames, Matplotlib for plots and scikit-learn for the algorithms and datasets, including the Iris dataset. The instructor recommends Visual Studio Code, though he says Anaconda also works. Some algorithms, such as gradient descent and KNN, are also encouraged to be written from scratch.

Does this course cover deep learning?

Only an introduction. The last session builds a neural network from logistic regression, with weights, sigmoid activations and hidden layers, and explains gradient-descent weight updates. The instructor says deep learning and the detailed maths, such as log loss, are left for advanced courses.

Last updated 18 Nov 2026 · Course length, chapters, views and dates from the YouTube videos on 18 Nov 2026 · Grow by Skill-Lync: free lessons from every course

Finished the course?

Ask about the certificate or your next course.

Leave your name and phone number. A Skill-Lync counsellor calls you back within one working day about the certificate for Machine Learning Fundamentals In Depth and the course that fits where you want to work.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Free full course · Skill-Lync

Start watching

Free, no payment. Leave your name and phone once and the whole Machine Learning Fundamentals In Depth course plays here, with every free lesson on Grow. A Skill-Lync counsellor may call you about the certificate or your next course.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.