Track 1 · Data Science Foundations
FoodHub Analysis — The Concepts Behind the Numbers
From raw data to business insights: the statistical and analytical techniques used in the FoodHub project.
Based on: FoodHub Data Analysis — Foundations for Data Science · Score 60/60
View Project →
Lesson 1
Exploratory Data Analysis (EDA)
Why we explore before we model. The three phases of EDA and the mindset behind them.
Lesson 2
Data Cleaning & Missing Values
Types of dirty data, what NaN really means, and why imputation can be dishonest.
Lesson 3
Descriptive Statistics
Mean, median, standard deviation, IQR — what they measure and when each lies to you.
Lesson 4
Outlier Detection — The IQR Method
What makes a value an outlier, Tukey's fences, and when to delete vs keep them.
Lesson 5
Data Visualisation
Histograms, boxplots, bar charts, pairplots — choosing the right chart for the question.
Lesson 6
Correlation & Multivariate Analysis
What correlation measures, why it doesn't imply causation, and how to read a heatmap.
Lesson 7
Bootstrap Confidence Intervals
Why a single average is dishonest, and how resampling gives you honest uncertainty ranges.
Lesson 8
Hypothesis Testing
The null hypothesis, p-values, t-tests — and why statistical significance ≠ practical significance.
Track 2 · Deep Learning
SVHN Digit Recognition — The Architecture of Vision
From a single artificial neuron to a 92% accurate convolutional network: the building blocks of deep learning.
Based on: SVHN Digit Recognition — Deep Learning Project · Best accuracy: 92.22%
View Project →
Lesson 1
Neural Networks
From a single artificial neuron to multi-layer networks — how machines represent and learn from data.
Lesson 2
Activation Functions
Why non-linearity is essential, ReLU vs LeakyReLU vs Softmax — and when to use each.
Lesson 3
How Neural Networks Learn
Loss functions, gradient descent, backpropagation, learning rate — the full training loop explained.
Lesson 4
Overfitting & Regularisation
Why memorising training data is useless, and how Dropout and BatchNorm force generalisation.
Lesson 5
Convolutional Neural Networks
Filters, feature maps, MaxPooling — why CNNs see images the way the visual cortex does.
Lesson 6
Model Evaluation
Confusion matrix, precision, recall, F1 — why accuracy alone tells you almost nothing.