Machine Learning in MATLAB

Overview

As part of the Machine Learning course at HTWG Konstanz, I worked on two hands-on projects involving unsupervised learning (clustering) and reinforcement learning.
The tasks were implemented and analyzed using MATLAB, focusing on applying machine learning techniques in real-world inspired scenarios with structured evaluations.

Unsupervised Learning: Clustering Animal Data

Using a dataset of 342 animals with 7 features each, I applied and compared k-means and DBSCAN clustering methods. The goal was to match clusters as closely as possible to the true animal classes.
→ Clustering Repository

Methods & Insights

• Data preprocessing: feature selection (columns 4–6), z-score normalization
• k-means: 3 clusters assumed, sqEuclidean distance, relabeling for evaluation
• DBSCAN: parameter tuning for epsilon and MinPts, noise filtering
• Evaluation: confusion matrix for both algorithms

Reinforcement Learning: Value Iteration & Policy Evaluation

A custom Markov Decision Process (MDP) with 12 states and 4 actions was implemented in MATLAB.
The task included computing the value function under a fixed policy (“always move up”), and using Value Iteration to derive the optimal policy.
→ Reinforcement Learning Repository

Methods & Implementation

• Created a grid-world-like MDP in MATLAB
• Implemented the Bellman equation for policy evaluation
• Developed a custom Value Iteration function
• Compared policies and value functions across iterations