CSC2042S 2025 Assignment 1: Unsupervised
Learning
In this assignment, you will apply K-means clustering to analyze global development patterns
using the World Development Indicators (WDI) dataset. The assignment will focus on the
practical challenges of applying clustering to real-world data, model optimization strategies, and
interpretation of results. In addition to submitting the code for reproducing your analysis, you
need to submit a report that documents your model design decisions and analyses the results.
Learning Objectives
Upon completion of this assignment, you should be able to:
• Handle real-world data preprocessing challenges in clustering applications
• Evaluate and optimize clustering algorithms through various initialization and
convergence strategies
• Assess cluster quality using multiple validation approaches
• Apply dimensionality reduction techniques to improve clustering performance
• Critically analyze the trade-offs in clustering design decisions
Dataset
The World Development Indicators database from the World Bank contains over 1,400 time
series indicators for 220 countries and territories from 1960 to present. The dataset contains key
development indicators such as:
• Economic indicators (GDP per capita, inflation, trade balance)
• Social indicators (life expectancy, literacy rate, population growth)
• Environmental indicators (CO2 emissions, forest area, renewable energy consumption)
• Infrastructure indicators (internet users, mobile subscriptions, electricity access)
The dataset is available on the course Amathuba site under Content / Resources / Datasets. The
source of the dataset is: https://datacatalog.worldbank.org/dataset/world-development-indicators.
Your code should load the dataset from a specified folder in the given format: Do not submit the
dataset with your assignment or let the code download the dataset. In the main file in the dataset,
WDICSV.csv, each row represents an indicator for one country, with columns for the value of
that indicator at different years. However, the indicator values are not always available for every
year. In your data analysis, treat each country-year combination as one multi-dimensional data
point, with development indicators serving as features. This will enable clustering countries with
similar indicators and to track the development indicators over time (e.g. how similar is South
Africa in 2024 to South Korea in 1990). Additionally, the Human Development Index
scores/classifications of countries are also included – this should only be used as a reference
classification for analyzing the final clustering.
