Data Science from Scratch
Learn how data becomes decisions: describing and cleaning data, the statistics behind it, and building and explaining machine learning models, with examples from retail, health, transport, media and finance.
Taught by Ayodele Odugbile, data and analytics professional and founder of Open Fraud Labs
What you'll learn
- Explain what data science is and how a project runs from question to answer
- Read a dataset's structure and identify data types correctly
- Summarise data with averages, spread and distributions
- Find outliers, handle missing data and clean messy datasets
- Choose the right chart and avoid common statistical traps
- Query data with SQL and work fluently with pandas
- Train, test and fairly evaluate machine learning models
- Explain model decisions and check them for bias
Skills you'll practise
Course outline
Six modules. Lessons unlock in order: watch the video, read the notes, solve the practice and pass the quiz to complete each one.
Foundations
- 01What is data science, really?
- 02Types of data, and why they matter
- 03Your first dataset: rows, columns & features
Statistics and exploring data
- 04Mean, median & mode: describing data with one number
- 05Spread: range, variance & standard deviation
- 06Distributions and the normal curve
- 07Outliers: errors, or the most interesting rows?
- 08Missing data and what to do about it
- 09Data cleaning basics
- 10Exploratory data analysis (EDA)
- 11Choosing the right chart
- 12Correlation is not causation
- 13Sampling and sampling bias
- 14Probability basics for data science
Tools of the trade
- 15Why Python for data science
- 16Meet the pandas DataFrame
- 17SQL basics: SELECT, WHERE, GROUP BY
Machine learning essentials
- 18Train/test split: why we hide data from the model
- 19Linear regression intuition
- 20Classification and logistic regression
- 21Overfitting vs underfitting
- 22Why accuracy can lie
- 23The confusion matrix, precision & recall
- 24Decision trees
- 25Feature engineering
- 26Cross-validation
- 27Imbalanced data: the fraud detection problem
Responsible data science and next steps
- 28Explainable AI: why did the model decide that?
- 29Data ethics and bias in models
- 30Your data science roadmap
Portfolio projects
- 31Project 1 · Part 1: Credit risk — framing the problem and preparing the data
- 32Project 1 · Part 2: Credit risk — what drives default?
- 33Project 1 · Part 3: Credit risk — a default model and an approval policy
- 34Project 2 · Part 1: Clinic no-shows — framing the problem and preparing the data
- 35Project 2 · Part 2: Clinic no-shows — who misses appointments, and why?
- 36Project 2 · Part 3: Clinic no-shows — predicting no-shows and targeting reminders
- 37Project 3 · Part 1: City rents — cleaning listings and building features
- 38Project 3 · Part 2: City rents — exploring prices across the city
- 39Project 3 · Part 3: City rents — a price model, error analysis and portfolio write-up
Portfolio projects
Three guided projects in different industries, each peer reviewed against a published rubric. You finish with work you can show employers.
Credit risk
Predict loan default and set a costed approval cut-off.
Clinic no-shows
Find who is likely to miss appointments and target reminders.
City rents
Model rents across a city and price new listings.
How you earn the certificate
- Complete every lessonWatch, read, practise and pass the quiz for all 30 lessons.
- Finish the three projectsFollow each one through its three parts and pass the quizzes.
- Pass your capstoneAnalyse a dataset of your choice; three learners review it with the rubric.
- Claim your certificateIt shows your registered name and an ID anyone can verify.
Your instructor
Ayodele Odugbile
Data and analytics professional and founder of Open Fraud Labs. Every lesson is built on real code and checked numbers, so what you see on screen is exactly what you'll get when you run it.