Principles and Techniques in Data Science

The objective of this course is to empower students to provide data-driven solutions for various problems. For this purpose, students will become familiar with the mathematical and statistical prerequisites for such approaches, learn the principles and steps of data-driven solutions including data analysis and visualization, statistical and probabilistic modeling, statistical inference, and decision-making under uncertainty. Through practical application of these methods to real-world problems, students will become familiar with the challenges of implementing these methods in practice.

1. Data Analysis
   • Introduction to the data science lifecycle
   • Data generation (questionnaires, census, controlled experiments)
   • Data collection and aggregation (data standardization, tabular data representation, filtering and aggregating data)
   • Data cleaning (outlier management, missing values, encoding and vector space representation)
   • Exploratory data analysis
   • Data visualization
   • Pattern recognition and hypothesis generation through data visualization
   • Understanding pitfalls in data analysis (data bias, insufficient features, confusing correlation with causation)
   • Hypothesis testing and p-value manipulation
2. Statistical Data Modeling
   • Introduction to modeling steps (cost function, parameter learning, prediction, decision theory)
   • Model generalization capability and its evaluation using cost functions
   • Training, validation, and test data separation
   • Overfitting, cross-validation, regularization
   • Optimization methods (gradient descent, Newton's method, momentum-based methods)
   • Probabilistic and Bayesian modeling
   • Statistical inference, model learning using estimation theory, prediction using trained models
   • Decision theory
   • Bias-variance tradeoff
   • Curse of dimensionality
3. Statistical Modeling in Practice
   • High-dimensional data visualization using t-SNE
   • Feature extraction and selection
   • Feature quantization using decision trees
   • Linear classification methods
   • Classification using decision trees
   • Classifier evaluation
4. Machine Learning Engineering in Production
   • Introduction to MLOps: end-to-end learning, continuous learning, data drift, concept drift, feature store, pipelines
   • Data lifecycle in production environments
   • Learning lifecycles and pipelines in production environments
   • Deployment of learning systems in production environments

• Exams: Midterm and final exams (50% of grade)
• Assignments and Project: Three theoretical assignments and one practical project to be submitted during the semester (50% of grade)