| Popis: |
Ovaj rad istražuje primjenu tehnika strojnog učenja za predviđanje ishoda NCAA košarkaškog turnira. Istraživanje se fokusira na tri popularna algoritma, točnije logističku regresiju, metodu potpornih vektora (SVM) i XGBoost, uspoređujući njihovu uspješnost u predviđanju rezultata turnira. Dodatno, koriste se tehnike grupiranja momčadi prema njihovim karakteristikama, s ciljem poboljšanja točnosti predviđanja ishoda turnira. Korišteni su povijesno podaci turnira i procjenjuje prediktivne sposobnosti modela kroz različite metrike performansi. Modeli trenirani prema statistikama momčadi uspoređeni su s modelima trenirani prema grupama momčadi. This thesis explores the application of machine learning techniques to predict the outcome of the NCAA basketball tournament. It is centered around three widely used algorithms, logistic regression, support vector machines (SVM), and XGBoost. The objective is to assess their effectiveness in forecasting tournament outcomes by comparing their performance. Additionally, team clustering techniques are employed to group teams based on their characteristics, aiming to enhance the accuracy of bracket predictions. The experiment utilizes historical tournament data and evaluates the models' predictive abilities through accuracy and bracket score. Models trained on classic features are compared to the ones trained on team clusters. |
