RedLectern topic
Learn how to frame questions, inspect data, reason with statistics, and assess predictive models.
For: Learners at any starting point, including people new to programming or statistics.
After this page you can: Plan a basic data investigation and explain what its evidence supports and leaves uncertain.
Page reviewed
Part of: Computer Science
Data science is applied work with data that uses computation, statistical ideas, and modeling to investigate questions. This hub covers preparing and inspecting data, quantifying uncertainty or relationships, and building or assessing predictive models. Machine learning is one possible modeling route within broader work with data.
A practical way to study is to begin with a question and a dataset, inspect and transform the data, use statistical methods to summarize or infer, then assess predictions with evidence set aside during model development. These steps connect data science to programming and machine learning, while leaving room for different methods and goals.
Start with MIT’s introductory computational thinking and data science course, which is intended for learners with little or no programming experience.
Start with: Introduction to Computational Thinking and Data Science
Use NIST’s exploratory data analysis chapter to revisit graphical and quantitative ways to examine a dataset.
Start with: Exploratory Data Analysis
Follow Google’s Machine Learning Crash Course modules on data, models, and evaluation; its modules are self-contained, so you can focus on a particular topic.
Start with: Machine Learning Crash Course
Begin by connecting the question you want to investigate to the examples and variables available in the dataset.
This lesson describes datasets as collections of examples and discusses data types, quality, and missing values.
Dataset
A collection of examples.
Feature
A variable that may be used as input to a model.
Suppose a table records a plant’s species and height. Name one possible example and one possible feature.
One example is a single plant’s row; height is one possible feature.
Exploratory data analysis uses techniques, mostly graphical, to reveal structure, notice anomalies, and check assumptions.
The chapter describes using graphical techniques to uncover structure, detect anomalies, test assumptions, and develop models.
Exploratory data analysis
An approach to examining data to gain insight into its structure and assumptions.
Graphical technique
A visual method for examining data, such as a plot.
A dataset has one unusually large value. Name one exploratory step you could take before deciding what to do with it.
Plot the values and inspect the unusually large observation in context before deciding whether it is an error or a meaningful value.
Statistical inference uses sample data to estimate or draw conclusions about a population, while acknowledging uncertainty in those estimates.
The lesson introduces populations, samples, and the distinction between population parameters and sample statistics.
Population parameter
A measure concerning a population.
Confidence interval
A range computed from sample statistics to estimate an unknown population parameter at a stated confidence level.
A sample survey estimates that 40% of respondents prefer option A. What does a confidence interval add to that estimate?
It gives a range of plausible values for the population proportion, using the sample data and a stated confidence level.
Modeling requires choosing how raw values are represented, and statistical or machine-learning methods can serve different questions.
Its modules introduce regression, classification, and ways to prepare numerical and categorical data.
Feature vector
An array of feature values representing one example for a model.
Feature engineering
Converting raw data into representations suitable for a model.
A model uses a person's age and a text category for an example. Name a data-preparation issue to consider.
The text category may need to be represented numerically before it can be included in the model’s feature vector.
Use validation data to guide model development and reserve test data for a final check on examples not used for training.
This lesson introduces residual analysis as a way to assess whether a simple linear regression model’s assumptions are reasonable.
Training set
The examples used to train a model.
Test set
A separate set of examples used to evaluate a trained model.
You repeatedly change a model after reviewing its test results. What is one risk?
You may start fitting the model to peculiarities of that test set, making it a weaker check of performance on new examples.
No recent updates are available.
Gareth James, Daniela Witten, Trevor Hastie, and Robert Tibshirani · 2021
Its broad method survey is paired with a chapter lab that demonstrates concepts in R or Python.
Ani Adhikari, John DeNero, and David Wagner · 2022
Developed for UC Berkeley’s Data 8, it combines programming and inference with analysis of real-world data.
Trevor Hastie, Robert Tibshirani, and Jerome Friedman · 2009
The second edition adds treatments of random forests, ensemble methods, and high-dimensional problems alongside its broad learning-method coverage.
Bin Yu and Rebecca L. Barter · 2024
Presents a framework for assessing the trustworthiness and relevance of data-driven results, with real-world case studies and code.
Foster Provost and Tom Fawcett · 2013
Its business framing complements the statistical methods and analysis-workflow books elsewhere in the list.
Wes McKinney · 2022
The third edition has an author-hosted open-access version and covers common Python data-analysis workflows.
Listed as Data Science courses: