Precedent-Based Adaptive Modelling Framework for Basketball Win-Prediction
(2025) DABN01 20251Department of Economics
Department of Statistics
- Abstract (Swedish)
- The empirical prediction of outcomes in competitive sports, particularly basketball, remains a complex challenge due to inherent game variability and the difficulty in discerning replicable strategic adaptations from random fluctuations, especially in the case of unexpected results ("upsets"). This thesis addresses this by introducing and evaluating novel features designed to enhance the predictive accuracy of NBA game outcomes. These features capture aggregate seasonal team strength markers and dynamic in-game performance indicators, building upon traditional box-score statistics.
Results demonstrate that pre-game models incorporating novel momentum features achieved 71.1% accuracy, surpassing Las Vegas betting lines (68.5%). For... (More) - The empirical prediction of outcomes in competitive sports, particularly basketball, remains a complex challenge due to inherent game variability and the difficulty in discerning replicable strategic adaptations from random fluctuations, especially in the case of unexpected results ("upsets"). This thesis addresses this by introducing and evaluating novel features designed to enhance the predictive accuracy of NBA game outcomes. These features capture aggregate seasonal team strength markers and dynamic in-game performance indicators, building upon traditional box-score statistics.
Results demonstrate that pre-game models incorporating novel momentum features achieved 71.1% accuracy, surpassing Las Vegas betting lines (68.5%). For in-game predictions, the GRU model, leveraging the sequential nature of quarterly data, demonstrated superior performance, achieving 83.69% accuracy and an AUC-ROC of 0.9157. This outperformed a strong ensemble model (82.59% accuracy, 0.9119 AUC) and highlighted the predictive power of cumulative in-game metrics. The findings confirm that the proposed novel descriptive and explanatory features significantly improve predictive accuracy, with sequential modeling offering particular advantages for capturing evolving game dynamics. This work contributes to understanding replicability in sports outcomes and provides a robust framework for both pre-game and adaptive in-game prediction. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9197631
- author
- Markovic, Dennis LU
- supervisor
-
- Simon Reese LU
- organization
- course
- DABN01 20251
- year
- 2025
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Tree Based Methods, Binary Classification, RNN, LightGBM, GRU
- language
- English
- id
- 9197631
- date added to LUP
- 2025-09-12 09:04:45
- date last changed
- 2025-09-12 09:04:45
@misc{9197631, abstract = {{The empirical prediction of outcomes in competitive sports, particularly basketball, remains a complex challenge due to inherent game variability and the difficulty in discerning replicable strategic adaptations from random fluctuations, especially in the case of unexpected results ("upsets"). This thesis addresses this by introducing and evaluating novel features designed to enhance the predictive accuracy of NBA game outcomes. These features capture aggregate seasonal team strength markers and dynamic in-game performance indicators, building upon traditional box-score statistics. Results demonstrate that pre-game models incorporating novel momentum features achieved 71.1% accuracy, surpassing Las Vegas betting lines (68.5%). For in-game predictions, the GRU model, leveraging the sequential nature of quarterly data, demonstrated superior performance, achieving 83.69% accuracy and an AUC-ROC of 0.9157. This outperformed a strong ensemble model (82.59% accuracy, 0.9119 AUC) and highlighted the predictive power of cumulative in-game metrics. The findings confirm that the proposed novel descriptive and explanatory features significantly improve predictive accuracy, with sequential modeling offering particular advantages for capturing evolving game dynamics. This work contributes to understanding replicability in sports outcomes and provides a robust framework for both pre-game and adaptive in-game prediction.}}, author = {{Markovic, Dennis}}, language = {{eng}}, note = {{Student Paper}}, title = {{Precedent-Based Adaptive Modelling Framework for Basketball Win-Prediction}}, year = {{2025}}, }