Introduction To Data Science

Explore how to map business problems to machine learning tasks by selecting classification or regression approaches, review common algorithms, and understand supervised learning versus unsupervised learning.

Explore naive Bayes, Bayes law, and conditional probability for text data with large feature spaces, using bag of words and 2 grams, smoothing, and Bernoulli modeling.

Learn to build a naive Bayes text classifier using two-gram features, smoothing, and log-probabilities, applied to Shakespeare vs Marlowe texts, with training and evaluation on Gutenberg data.

Explore the background and theory of linear regression, including estimating a numeric outcome from predictors with least squares, and understand key concepts like unbiased predictions, residuals, and Gauss-Markov.

Apply linear regression to housing data, build models to predict price, handle categorical variables and interactions, and validate with hold-out cross-validation while refining features to improve rmse on 73 houses.

Explore logistic regression background as a method to estimate probabilities with a sigmoid, model the log odds, and minimize deviance for two-class classification while diagnosing potential issues and applying regularization.

Learn to fit a logistic regression model (binomial GLM with logit link) to predict abalone with ten or more rings using noninvasive measurements, then evaluate accuracy and deviance.

Train a random forest on a letter recognition dataset, assess performance with out-of-bag estimates and a test, and examine variable importances to select top features.

Explore generalized additive models (GAM) to model nonlinear variable effects in both regression and classification, using splines to reshape predictors and residuals, and improve fit.

Explore support vector machines, including linear SVM and kernel tricks with gaussian and cosine kernels, to model complex boundaries by maximizing margin and using support vectors for classification and regression.

learn gradient boosting to fit complex models by sequentially refitting decision-tree sub learners to residuals or pseudo residuals for classification and regression, with shrinkage and depth controlling regularisation.

Explore how regularisation improves linear and logistic regression by applying lasso and elastic net penalties to reduce overfitting, handle collinearity, and enable feature selection through cross-validated lambda.

Explore evaluating regression and classification models using metrics like RMSE, adjusted R-squared, deviance, ROC AUC, and gain curves, with true-versus-prediction plots.

Learn how to load data into R from multiple sources—typed data, built-in datasets, csv/tsv files, Excel via a data package, and databases—preparing data for analysis.

Visualize data to explore distributions and relationships using graphs. Learn to use clear, informative visualizations: histograms, density plots, scatter plots, and trellis plots for analysis and communication.

Identify missing values in data and learn how to handle them in training and testing of predictive models, using a unified representation for missing values and imputation or indicator methods.

Explore the shape of data by mastering wide and long forms, pivoting, reshaping, aggregation, and joining to normalize and analyze data across rows, columns, and groups.

Explore how to handle categorical variables in data science by encoding them as indicators or dummy variables, using model matrices, respecting reference level, and avoiding unsafe numeric encodings.

Learn practical data transformations to fix data problems, reduce outliers, and improve model robustness using log, square root, inverse, arc-sinh, and cdf-based transforms, as well as winsorizing and PCA considerations.

Explore advanced topics in data science, including privacy, governance, and ethics. Discuss reproducible research, data provenance, big data, data engineering, and machine learning challenges.

Practice on your own by selecting a dataset, asking questions, and building a model to predict a fact, using datasets from the UCI repository and the Journal of Statistics Education.