Csaba Gáspár, assistant lecturer (TMIT)

István Nagy-Rácz, technical assistant (TMIT)

Kötelező:

The key issue in the utilization of machine learning and data science knowledge is that we can properly map a real data set and an actual business problem to our machine learning and data analysis toolset. The aim of the subject is to give students a deeper practical experience in this data analysis process by solving several real case studies to demonstrate the order and manner in which the methods should be used.

1. Introduction, methods, technologies, overview of used programming languages and technologies (e.g. Python) - Tabular data management, DataFrame-based problem solving, questions of code efficiency

2. Supervised learning - Advanced regression methods over a real estate data set, advanced methods in data preparation, management of time trends, special data preparation methods and their impact on forecasting

3. Supervised learning – Advanced classification task for solving a credit assessment task

4. Supervised learning – Complex classification evaluation methods, unique objective functions, optimization in case of unique objective functions. The special properties of the ROC curve and AUC value, the relationship between error detection and evaluation functions, the evolution of evaluation in the light of business needs

5. Clustering procedures - Challenges of customer segmentation based on clustering, data preparation difficulties, selection of clustering methods, explainability of clustering results, recognition of trivial clustering situations, story-telling related to clustering

6. The relationship between story telling and the interpretability of models, algorithmic issues of explainability of models

7. Anomaly detection – Solving a complex anomaly detection task over a time-varying data set

8. Anomaly detection – Combining anomaly score values, incorporating a feedback process into the entire analysis sequence

9. Advanced methods of variable generation, their relationship with variable selection methods - Presentation of the methods and challenges of variable selection, the feature engineering process that utilizes the results of the selection

10. Description of the large homework task, preparation of its initial solution, description of the pitfalls of the data analysis task

 

- Assignments (small homeworks) to be submitted within the Moodle system (5 pieces) - assignments are issued every two weeks, two weeks are available to complete the assignment.

- Participation in the data science competition (large homework) related to the subject, where the baseline specified during the semester must be reached.

Small homework scores count for 40%, while the data science competition scores count for 60% of the grade. At the competition, a baseline level must be reached in order for the student to receive at least a pass (2) grade, this level is recorded when the competition is announced. During the competition, a supervised learning task must be solved, the exact metric is determined by the competition task announced in the given semester. The students start the competition independently, the available score depends on the results achieved by all the students participating in the competition in the given year.

At least 40% of the total score (small homeworks plus competition) must be achieved to receive credit.

Assignments (small homeworks) to be submitted can be made up 2 weeks after the submission deadline. If this deadline extends beyond the teaching period, they must be submitted in one week after the teaching period.

The submission of the large homework (competition) is continuous from the middle of the semester, the submission cannot be replaced here, the competition closes at the end of the teaching period.