logo
COURSE INFORMATION
Course Title: MACHINE LEARNING AND AI IN FINANCE
Code Course Type Regular Semester Theory Practice Lab Credits ECTS
FINT 404 B 2 3 0 0 3 7.5
Academic staff member responsible for the design of the course syllabus (name, surname, academic title/scientific degree, email address and signature) Dr. Florenc Skuka fskuka@epoka.edu.al
Main Course Lecturer (name, surname, academic title/scientific degree, email address and signature) and Office Hours: Dr. Florenc Skuka fskuka@epoka.edu.al
Second Course Lecturer(s) (name, surname, academic title/scientific degree, email address and signature) and Office Hours: M.Sc. Ferit Melih Akaybicen fakaybicen@epoka.edu.al
Language: English
Compulsory/Elective: Compulsory
Study program: (the study for which this course is offered) Master of Science in FinTech
Classroom and Meeting Time:
Teaching Assistant(s) and Office Hours: NA
Code of Ethics: Code of Ethics of EPOKA University
Regulation of EPOKA University "On Student Discipline"
Attendance Requirement: Minimum 75% attendance required. Active participation in lab sessions and the final competition is mandatory.
Course Description: -
Course Objectives: This course introduces fundamental machine learning concepts and their applications in finance through a hands-on, project-based approach. Students build ML models in Python that control agents inside a 3D web simulation, mapping each week's ML topic to a concrete project milestone. Topics include linear and multivariate regression, regularization, clustering, classification, decision trees, ensemble methods (Random Forests, XGBoost), neural networks (forward pass and backpropagation), convolutional neural networks, ML pipelines, and anomaly detection. The course culminates in a live autonomous agent competition integrating all techniques learned.
BASIC CONCEPTS OF THE COURSE
1 Machine Learning: A field of computer science that gives systems the ability to learn from data and improve performance on tasks without being explicitly programmed for each case.
2 Supervised Learning: A learning paradigm where models are trained on labeled data (input-output pairs) to learn a mapping function, used for regression (continuous outputs) and classification (categorical outputs).
3 Unsupervised Learning: A learning paradigm where models discover hidden patterns and structure in unlabeled data, including clustering (K-Means, DBSCAN) and dimensionality reduction (PCA, t-SNE).
4 Gradient Descent: An iterative optimization algorithm that minimizes a cost function by updating model parameters in the direction of the negative gradient, foundational to training regression models and neural networks.
5 Overfitting and Regularization: Overfitting occurs when a model memorizes training data noise rather than learning general patterns. Regularization (Ridge L2, Lasso L1) adds penalty terms to constrain model complexity.
6 Classification: The task of predicting categorical labels for input data. Evaluated using confusion matrices, precision, recall, and F1 score. Includes logistic regression, decision trees, and SVMs.
7 Ensemble Methods: Techniques that combine multiple models to improve prediction accuracy. Includes bagging (Random Forests), boosting (XGBoost/Gradient Boosting), and stacking.
8 Neural Networks: Computational models inspired by biological neurons, consisting of layers of interconnected nodes with learnable weights. Trained via backpropagation using the chain rule of calculus.
9 Convolutional Neural Networks (CNNs): Specialized neural networks that use convolution operations to exploit spatial structure in data (images, grids), consisting of convolutional layers, pooling layers, and fully connected layers.
10 Anomaly Detection: The identification of data points that deviate significantly from normal behavior. Uses techniques like multivariate Gaussian models and one-class SVM, applied in fraud detection, system monitoring, and quality control.
COURSE OUTLINE
Week Topics
1 Introduction to Machine Learning: What is ML? Supervised vs. unsupervised vs. reinforcement learning. Features, labels, datasets, train/test split. The ML workflow: collect, explore, model, evaluate, deploy. Setup of the 3D simulation environment (Three.js + Rapier.js). Python SDK connection, sensor data collection, and exploratory data analysis.
2 Linear Regression (Single Variable): Simple linear regression (y = wx + b). Cost function (MSE), gradient descent intuition. Fitting a line, interpreting slope and intercept. Residual analysis, R-squared score. Assignment: Predict projectile landing distance from launch angle using the simulation.
3 Multivariate Regression: Multiple linear regression with feature matrices. Normal equation, vectorized gradient descent. Feature scaling (standardization, min-max). Interpreting coefficients and feature importance. Train/test split and evaluating generalization. Assignment: Predict 3D landing position from multiple features (angle, force, wind).
4 Regularization and Learning Curves: Polynomial feature expansion. Overfitting vs. underfitting (bias-variance tradeoff). Ridge regression (L2) and Lasso regression (L1). Regularization parameter selection. Learning curves and cross-validation (k-fold). Assignment: Tame overfitting with regularized polynomial models.
5 Clustering and Unsupervised Learning: K-Means algorithm (initialization, iteration, convergence). DBSCAN (density-based clustering). Elbow method, silhouette score. Dimensionality reduction (PCA, t-SNE). Assignment: Terrain map segmentation — discover terrain types from unlabeled sensor data without ground truth labels.
6 Linear Classification: Classification vs. regression. Logistic regression (sigmoid function, log-loss). Decision boundaries. Multi-class classification (one-vs-rest, softmax). Confusion matrix, precision, recall, F1 score. Assignment: Terrain classification from sensor data — binary and multi-class classifiers.
7 Decision Trees and Random Forests: Decision tree splits, Gini impurity, entropy. Tree depth, pruning, overfitting control. Random forests: bagging, feature subsampling. Feature importance and out-of-bag error estimation. Assignment: Tree-based terrain classification and predicting terrain ahead of the agent.
8 Ensemble Methods and XGBoost: Gradient boosting (sequential error correction). XGBoost/LightGBM practical usage. Bagging vs. boosting vs. stacking. GridSearchCV for hyperparameter tuning. SVM overview (max margin, kernel trick). Assignment: Path safety classifier and comprehensive model comparison shootout.
9 Neural Networks — Forward Pass: Neurons, layers, activation functions (ReLU, sigmoid, tanh). Weight matrices, bias vectors. Forward propagation as matrix multiplication + activation. Network architecture design. Assignment: Implement forward pass from scratch in NumPy, deploy a pre-trained navigation network to autonomously drive the agent.
10 Neural Networks — Backpropagation: Backpropagation algorithm (chain rule). Loss functions (MSE, cross-entropy). Training loop: forward, loss, backward, update. SGD, mini-batch, momentum, Adam optimizer. Behavioral cloning from expert demonstrations. Assignment: Train a navigation network from scratch via behavioral cloning.
11 Convolutional Neural Networks: Convolution operation (filters, stride, padding). Feature maps and learned filters. Pooling layers (max pooling). CNN architectures: Conv-ReLU-Pool-FC. Spatial structure in data. Assignment: Train a CNN on 32x32 grid observations for agent navigation, compare with fully connected networks.
12 ML Pipelines, Feature Engineering, and Modern AI: sklearn Pipeline and ColumnTransformer. Feature engineering (rolling means, deltas, interaction features). Handling missing data, outliers, scaling. Experiment tracking. Modern AI landscape: transformers, LLMs, reinforcement learning, RLHF. Assignment: Build a competition-ready pipeline.
13 Anomaly Detection: Multivariate Gaussian distribution fitting. Anomaly scoring (Mahalanobis distance, negative log-likelihood). One-class SVM. Threshold selection using F1 score. Precision-recall tradeoff for imbalanced problems. Assignment: Detect agent malfunctions and terrain anomalies in the simulation.
14 Competition and Presentations: System integration — combining multiple ML models into an autonomous agent. Model deployment and real-time inference. Live multiplayer tournament (Time Trial, Target Practice, Anomaly Survival, Combined rounds). 10-minute technical presentations covering approach, architecture, results, and lessons learned.
Prerequisite(s): None
Textbook(s): Aurélien Géron, "Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow", 3rd Edition, O'Reilly Media, 2022.
Additional Literature: 1. Sebastian Raschka, "Machine Learning with PyTorch and Scikit-Learn", Packt Publishing, 2022. 2. Jake VanderPlas, "Python Data Science Handbook", 2nd Edition, O'Reilly Media, 2023. 3. Marcos López de Prado, "Advances in Financial Machine Learning", Wiley, 2018. 4. scikit-learn documentation: https://scikit-learn.org/stable/ 5. PyTorch Tutorials: https://pytorch.org/tutorials/
Laboratory Work: Weekly hands-on lab sessions where students build and deploy ML models in Python, interact with the 3D simulation environment via REST/WebSocket APIs, and test their models in real-time.
Computer Usage: Extensive use of Python (NumPy, pandas, scikit-learn, PyTorch, matplotlib), Jupyter Notebooks, Node.js relay server, and a custom 3D web simulation (Three.js + Rapier.js physics engine) running in Chrome browser.
Others: No
COURSE LEARNING OUTCOMES
1 Explain fundamental ML concepts (supervised, unsupervised, reinforcement learning) and apply the ML workflow (collect, explore, model, evaluate, deploy) to real-world problems.
2 Implement linear and multivariate regression models from scratch, perform feature engineering and scaling, and interpret model coefficients in the context of financial and physical systems.
3 Apply regularization techniques (Ridge, Lasso) to control overfitting, diagnose bias-variance tradeoff using learning curves, and select hyperparameters via cross-validation.
4 Apply unsupervised learning algorithms (K-Means, DBSCAN) for clustering and segmentation, and evaluate results using elbow method, silhouette score, and dimensionality reduction (PCA, t-SNE).
5 Build and evaluate classification models (logistic regression, decision trees, random forests, XGBoost, SVM) using confusion matrices, precision, recall, F1 score, and systematic model comparison.
6 Implement neural network forward and backward passes from scratch in NumPy, understand backpropagation, and train networks using behavioral cloning for autonomous agent control.
7 Design and train convolutional neural networks (CNNs) in PyTorch for spatial data processing, and compare CNN architectures with fully connected networks on structured observation data.
8 Construct reproducible ML pipelines using sklearn Pipeline and ColumnTransformer, perform systematic feature engineering, and track experiments for model comparison.
9 Build anomaly detection systems using multivariate Gaussian models and one-class SVM, select thresholds via precision-recall analysis, and apply these techniques to detect system malfunctions.
10 Integrate multiple ML models into a complete autonomous system, deploy models for real-time inference via REST/WebSocket APIs, and communicate technical results through presentations and portfolios.
COURSE CONTRIBUTION TO... PROGRAM COMPETENCIES
(Blank : no contribution, 1: least contribution ... 5: highest contribution)
No Program Competencies Cont.
Master of Science in FinTech Program
1 Graduates will master the fundamentals of fintech, including technological innovations, digital financial services, and disruptive business models, and apply this knowledge to create fintech solutions. 3
2 Graduates will be proficient in blockchain technologies and cryptocurrency, understanding their applications in finance, such as digital currencies, smart contracts, and decentralized finance (DeFi) systems. 2
3 Graduates will gain expertise in digital banking and innovation, including the ability to design and manage digital banking solutions, improving customer experience and operational efficiency. 2
4 Graduates will demonstrate a deep understanding of corporate finance principles, enabling them to analyze complex financial scenarios and make informed strategic decisions. 4
5 Graduates will have the skills to thoroughly analyze financial statements and assess the financial health and performance of organizations, aiding in investment and decision-making processes. 5
6 Graduates will acquire advanced skills in big data analytics, enabling them to extract valuable insights from large datasets to drive data-driven financial strategies and decisions. 3
7 Graduates will apply machine learning and artificial intelligence techniques to analyze financial data, automate decision-making processes, and enhance risk management in the financial sector. 1
8 Graduates will gain knowledge in cybersecurity resilience, enabling them to understand cyber threats in financial systems and data and effectively cooperate in mitigating fraud risks in fintech operations. 4
9 Graduates will develop a strong ethical foundation and be equipped to navigate corporate social responsibility and sustainability challenges in the fintech industry, promoting responsible and sustainable financial practices. 4
10 Graduates will be capable of applying interdisciplinary knowledge to address complex challenges at the intersection of finance and technology, fostering innovation and adaptability in a rapidly evolving fintech landscape. 5
COURSE EVALUATION METHOD
Method Quantity Percentage
Homework
10
3
Project
1
35
Final Exam
1
35
Attendance
0
Total Percent: 100%
ECTS (ALLOCATED BASED ON STUDENT WORKLOAD)
Activities Quantity Duration(Hours) Total Workload(Hours)
Course Duration (Including the exam week: 16x Total course hours) 16 4 64
Hours for off-the-classroom study (Pre-study, practice) 14 4 56
Mid-terms 0
Assignments 10 3 30
Final examination 1 20 20
Other 1 30 30
Total Work Load:
200
Total Work Load/25(h):
8
ECTS Credit of the Course:
7.5
CONCLUDING REMARKS BY THE COURSE LECTURER

This course is designed around a core philosophy: machine learning is best learned by building and deploying real systems. Rather than studying algorithms in isolation, students apply every ML technique to control autonomous agents inside a live 3D simulation, watching their models succeed or fail in real time. The progression from simple linear regression to neural networks and anomaly detection mirrors the complexity ladder of real-world ML engineering. By the final week's competition, students have not just learned individual algorithms — they have integrated them into a complete, deployable system. This project-based approach builds the practical skills, debugging intuition, and system-level thinking that define effective ML practitioners in the FinTech industry.
2014 - 2026 © Epoka University, ICTC Office