Machine Learning Classification Bootcamp in Python

Explore how machine learning, a subset of artificial intelligence, learns from training data to make predictions and decisions, covering supervised, unsupervised, and reinforcement learning, including classification and clustering.

This lecture explains supervised, unsupervised, and reinforcement learning with image classification examples, introduces deep learning as a subset of machine learning, and previews key classification techniques in Python with scikit-learn.

download and install the anaconda distribution across Windows, Mac OS, and Linux, including Jupyter notebooks and Python 3.7 for data science and machine learning.

Learn to use Jupyter Notebooks, an open source web app that blends live code, text, equations, and visualizations for data analysis and machine learning workflows.

Learn how to launch and use Jupyter notebooks with Anaconda, write and run code cells, mix markdown and code, visualize results, and manage notebooks for machine learning projects.

Explore the theory and intuition of logistic regression classifiers, build two from-scratch projects, and apply sigmoid probabilities, pandas, seaborn, and scikit-learn to real-world binary prediction tasks.

Explore logistic regression intuition for binary classification, contrasting with linear regression; learn to use a sigmoid probability and a threshold to classify data, with two practical projects.

Explore the confusion matrix, including true positives, true negatives, false positives, false negatives, and type I and II errors, and learn to compute accuracy in classification.

Train a logistic regression classifier on the Titanic dataset to predict survival using features like age, P class, sex, fare, embarked, and family-related counts.

Load the Titanic dataset with pandas, build a logistic regression classifier, and visualize and clean data by dropping noninformative columns, handling NaN, and inspecting using head and tail.

Visualize and explore a titanic-like dataset using pandas, seaborn, and matplotlib. Inspect survival patterns across pclass, embarked, sex, age, fare, siblings, and parch to prepare for a logistic regression project.

Prepare the data for training a logistic regression classifier by cleaning missing values and dropping unused columns, then fill age using a sex-based simple function.

Learn to clean and prepare the training dataset by dropping unnecessary columns, handling missing values, and encoding sex with get dummies to build X and y for training.

Split the data into 80/20 training and testing sets, then train a logistic regression classifier with sklearn using X_train and Y_train, preparing for evaluation.

Apply logistic regression to evaluate a trained classifier, generate predictions on test data, and assess performance with a confusion matrix and classification report, achieving around 81–82% accuracy.

Apply logistic regression to predict whether a Facebook ad will be clicked, using clean data with time spent on site and salary, building and evaluating a classifier with visualizations.

Apply logistic regression to a binary ad-click prediction problem, import libraries, load the Facebook ads dataset from CSV with pandas, handle encoding, and inspect the first and last rows.

Visualize and explore a dataset for predicting ad clicks by separating data into clicked and not clicked, and plot time spent on site against salary with scatter plots and histograms.

Drop emails, names, and country to clean data; define X with time spent on site and salary, and Y with clicked. Scale features with a standard scaler to train X.

Train a logistic regression model on an 80/20 split using time spent and salary to predict clicks, then evaluate with confusion matrix and classification report, reaching about 86% test accuracy.

Visualize and evaluate a logistic regression classifier on training and testing data, using a boundary diagram to show model predictions versus true points, highlight misclassifications, and relate results to the confusion matrix with an accuracy around 86%.

Apply support vector machines to classify retirement eligibility and tumor benign vs malignant, optimize the maximum margin hyperplane, and tune C and gamma with grid search.

Explore the intuition behind support vector machines and the maximum margin hyperplane. Learn to apply SVM from scratch with training and testing, using two features and real-world examples.

Apply support vector machines to classify tumors as malignant or benign using 30 features such as radius, texture, perimeter, and area from the Wisconsin Medical Center dataset.

Apply support vector machines to classify breast cancer as malignant or benign using a 569-sample, 30-feature dataset (radius, texture, perimeter, area), import data, visualize features, and train the model.

Visualize cancer data with seaborn pair plots and a correlation heatmap to reveal class separation. Focus on mean radius, texture, area, perimeter, and smoothness.

Prepare X and Y from the cancer data, split with 0.25 test size using train_test_split, and train a support vector classifier.

Test the trained SVC model on the 25% test set, generate predictions, and evaluate with a confusion matrix and classification report; 45% accuracy prompts exploring a 35/65 split.

In machine learning classification bootcamp, improve the support vector classifier by normalizing and scaling data to 0–1, applying linearization, and training on scaled data for high accuracy.

Improve a support vector classifier through normalization and grid search optimization of the C and gamma parameters using an RBF kernel to balance generalization and overfitting.

Apply support vector machines to bank retirement predictions by importing the dataset, visualizing age and 401k savings, and preparing for model training and evaluation with confusion matrices and classification reports.

Drop the customer ID and train a support vector classifier using age and 4 to 1 savings as inputs, with retire as the target, then apply train-test split.

Apply feature scaling to improve SVM classification results by normalizing data to 0 to 1, train the SVM model, and evaluate with a confusion matrix and a classification report.

Enhance support vector machines with feature scaling and grid search over c and gamma with an rbf kernel, achieving 94% precision on bank retirement predictions, up from 92%.

Master k-nearest neighbors by exploring theory and building two practical projects from scratch, then train models on training data with Euclidean distance and plot decision boundaries.

Learn the intuition of the k-nearest neighbors algorithm for classification, using Euclidean distance to pick the k closest points and a majority vote to classify new data.

Apply k-nearest neighbors to classify t-shirt size (small or large) using height and weight data, with training and testing splits, distance calculations, and a hands-on coding walkthrough in Jupyter notebooks.

Import libraries and the t shirt sizing dataset, choose height and weight as features, apply label encoding, and perform a train-test split for knn.

Train a k-nearest neighbors classifier (n_neighbors=5, p=2 Minkowski) on label-encoded training data and test with X_test and Y_test. Evaluate results via a confusion matrix and heatmap.

Visualize train and test data with a k-nearest neighbors classifier, showing predicted regions and correctly classified points. Explore the nonlinear decision boundary and the role of height and weight features.

Apply k nearest neighbors to classify iris species using sepal length, sepal width, petal length, and petal width, and evaluate with a confusion matrix and a classification report.

Apply k-nearest neighbors to classify iris species by visualizing sepal and petal features. Load data, explore with head and tail, and create scatter, violin, and pair plots, plus correlation heatmaps.

Train a k-nearest neighbors classifier on the iris dataset by preparing four features, splitting data into training and testing sets, and fitting the model with k=5 and euclidean distance.

Evaluate a trained knn classifier on the iris data using y_predict and X_test, visualize the confusion matrix, and review the classification report for accuracy and neighbor impact.

Master decision trees and random forest classifiers in supervised learning, understand decision nodes and leaves, and apply sentiment analysis on Amazon reviews and kyphosis classification using ensemble methods in Python.

Explore how decision trees split data with thresholds to classify outcomes, using decision nodes and leaves, and see how random forests extend this intuition in supervised learning.

Understand how random forest combines multiple decision trees into an ensemble that uses majority voting to improve classification accuracy while reducing overfitting through random data sampling.

Apply a random forest classifier to Amazon Alexa review data to perform sentiment analysis. Explore data, perform cleaning and feature engineering, train and evaluate the model, and visualize confusion matrix.

Import the Amazon Alexa dataset and build a mini random forest classifier to predict positive or negative reviews using features like rating, date, variation, and the review text.

Visualize product feedback and variations using count plots, histograms, and bar plots to compare positive versus negative reviews and rating distributions.

Clean the data by dropping date and rating columns, encode variations with get_dummies to produce a binary target, and prepare features for classifier training.

Transform verified reviews into numeric features with CountVectorizer, creating a high-dimensional feature matrix for sentiment classification, and prepare X and y for model training.

Train a random forest classifier on the dataset, perform a train/test split, and evaluate with a confusion matrix and classification report.

Evaluate a random forest classifier on training and testing data, visualize confusion matrices, and inspect the classification report to assess accuracy and optimize the number of trees.

Improve your random forest classifier for Amazon Alexa reviews by adding count vectorizer translations and review length as features, then evaluate with train-test split, confusion matrix, and classification report.

Build and evaluate classification models using decision trees and random forest on the kyphosis dataset, with train-test splits, confusion matrices, and input features like age, vertebrae, and start.

Explore the kyphosis classification project in Python, loading the dataset, encoding targets, visualizing distributions and correlations, and prepping data for decision trees and random forest.

Prepare and clean the kyphosis data, split into training and testing sets, and train a decision tree while identifying feature importances, with start ranked highest. Compare with a random forest.

Train and evaluate a decision tree classifier using training and testing splits, visualize and interpret confusion matrices and classification reports, and prepare to boost performance with a random forest classifier.

Compare decision trees and random forest classifiers to boost accuracy from 82% to nearly 90% by using a random forest with 150 estimators, evaluating with confusion matrices and classification reports.

Explore naive Bayes theory and learning outcomes, including Bayes theorem, prior, likelihood, and marginal likelihood, and build SMS spam and credit card fraud detectors with scikit-learn and CountVectorizer.

Explore the intuition of naive bayes classification through a two-class bank example using age and 401k savings, and learn how prior, likelihood, and posterior probabilities drive decisions.

Explore the Naive Bayes math behind classifying retiree status using age and savings, deriving prior, likelihood, and posterior, and applying the marginal likelihood for normalization.

Train a Naive Bayes classifier to detect spam versus ham in emails and SMS messages by loading data, visualizing it, applying a count vectorizer, and evaluating the model.

Visualize and analyze a spam vs ham dataset by splitting data, computing percentages, and plotting counts with seaborn, while preparing for training and feature engineering.

Apply count vectorizer from sklearn to translate text into a numeric feature matrix, then inspect feature names and the resulting array on a spam and ham dataset.

Train a multinomial naive Bayes classifier on a countvectorizer spam/ham dataset (5700 samples, 37,000 columns), then transform test messages and predict spam or ham to validate performance.

Train a Naive Bayes classifier with a train-test split of 80/20, using a count vectorizer in scikit-learn, and prepare for evaluation in the next lecture.

Explore credit card fraud detection using a Naive Bayes classifier on an unbalanced PCA-transformed dataset with V1–V28 features, training/testing splits, and model evaluation through confusion matrices.

Import and explore a credit card transaction dataset in Python, load data, separate fraud and non-fraud samples, visualize distributions, kernel density estimates, and correlations for model training.

Apply feature scaling to the amount column with a standard scaler, create amount_norm, and prepare X and y for a train–test split to train the model.

Train a Gaussian naive Bayes classifier on the training data, evaluate with a confusion matrix and classification report, and note 24 misclassified test samples in an unbalanced dataset.

Improve the Gaussian Naive Bayes classifier by dropping non-informative features, retraining on a reduced feature set, and evaluating with confusion matrices and recall improvements for fraud detection.