METU
Computer Engineering Department
CENG 562 – Machine Learning
Project Document
Objective:
The term project is a central component of this course. It offers you the opportunity to apply
machine learning techniques to a problem of your choice, conduct independent exploration,
and gain experience in research-style investigation and implementation.
Note: Projects must focus on non-deep learning machine learning methods. You are
expected to apply, modify, or analyze classical ML techniques such as decision trees,
ensemble methods, kernel methods, probabilistic models, graphical models, or optimization-
based approaches.
Group Work Requirement:
All term projects will be completed in groups of exactly 3 students.
You are encouraged to collaborate actively and divide work fairly among group members.
Each student must understand and be able to explain all parts of the project.
Project Goals:
By the end of the project, you will demonstrate your ability to:
Formulate and scope a research or application-driven ML problem.
Conduct a structured literature review.
Design, implement, and evaluate classical ML models.
Compare methods using appropriate datasets and metrics.
Communicate your work in both written and oral formats.
Timeline:
Milestone Description Due Date
Group Formation & Submit names of all 3 group members and a 1-page
19.10.2025
Topic Proposal project proposal.
3–4 pages summarizing the problem, literature,
Progress Report 30.11.2025
methods tried, and preliminary results.
Final Report & Code Final report (6–10 pages) and complete codebase. 04.01.2026
Present your project during class or a dedicated
Presentation 17.12.2025
session.
FNA
METU
Computer Engineering Department
CENG 562 – Machine Learning
Project Types (No Deep Learning)
You may choose one of the following types of projects or propose your own:
1. Algorithmic Projects
Modify or extend an existing classical ML algorithm.
Study theoretical properties: convergence, complexity, robustness.
Compare regularization techniques across models.
2. Applied Projects
Apply classical ML to a real-world problem: economics, education, biology, social
science, etc.
Analyze the limitations of existing models in specific domains.
Develop interpretable models for decision-making.
3. Empirical Comparison / Benchmarking
Compare multiple classical ML methods on one or more datasets.
Evaluate generalization, bias-variance trade-off, robustness to noise, etc.
Study how data preprocessing affects model performance.
4. Theoretical / Analytical Work
Analyze properties of kernel methods or ensemble techniques.
Explore sparsity, convexity, or duality in ML models.
Theoretical analysis of model complexity and generalization bounds.
Sample Project Ideas
Interpretability & Simplicity
Compare model interpretability of decision trees, rule-based systems, and logistic
regression.
Apply interpretable models to sensitive domains (e.g., credit scoring or health data).
Probabilistic & Graphical Models
Build and infer in Bayesian networks or Hidden Markov Models (HMMs).
Compare maximum likelihood and Bayesian estimation for probabilistic models.
FNA
METU
Computer Engineering Department
CENG 562 – Machine Learning
Kernel Methods
Compare SVMs with different kernels on real-world structured data.
Explore kernel PCA vs. standard PCA on high-dimensional data.
Model Selection & Evaluation
Study cross-validation strategies (e.g., nested CV, time-series CV).
Evaluate the effect of hyperparameter tuning under data constraints.
Time Series & Structured Data
Apply classical models like ARIMA, Gaussian Processes, or Kalman filters to time-
series data.
Compare probabilistic forecasting methods without using RNNs or transformers.
Optimization & Regularization
Study convex optimization in ML: coordinate descent, stochastic gradient, etc.
Explore regularization (L1, L2, ElasticNet) and their effect on model sparsity and
performance.
Final Deliverables
1. Final Report (6–10 pages, in academic format).
o Abstract
o Introduction and motivation
o Related work
o Methodology
o Experiments and evaluation
o Conclusion and future work
o References
2. Code Repository
o Public or private GitHub repo with clear documentation.
o Should include reproducible scripts or notebooks.
3. Presentation
o Short oral presentation (~8–10 minutes) or poster session summarizing your
work.
You will be given the template of each deliverable seperately.
Academic Integrity
Projects must be your group’s original work.
Use of existing code, libraries, or datasets is allowed with proper citation.
Plagiarism or code copying will result in strict academic penalties.
FNA
METU
Computer Engineering Department
CENG 562 – Machine Learning
Need Help?
You’re encouraged to seek help on the course discussion platform for:
Feedback on your idea or scope.
Guidance with implementation or troubleshooting.
Questions about literature or datasets.
Good luck, and enjoy your exploration of machine learning!
FNA