Forecasting sport-matches through data mining II
In recent years, machine learning, particularly deep learning has become in- creasingly studied, and applied in different areas of interest. One of them is the prediction of sports game results. Many have attempted to develop models to predict results of National Col- legiate Athletic Associatio...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/70198 |
| _version_ | 1811693592424480768 |
|---|---|
| author | Fu, Jiaxiang |
| author2 | Pan Jialin, Sinno |
| author_facet | Pan Jialin, Sinno Fu, Jiaxiang |
| author_sort | Fu, Jiaxiang |
| collection | NTU |
| description | In recent years, machine learning, particularly deep learning has become in-
creasingly studied, and applied in different areas of interest. One of them is
the prediction of sports game results.
Many have attempted to develop models to predict results of National Col-
legiate Athletic Association (NCAA) Men's Basketball Tournament. However,
existing models require manual e ort to rst extract features from data before
they can be trained to make predictions. These manual processes typically
need to be repeated when model is applied to new data, a new season of game,
for example.
This project aims to develop a model that can generalize its training to
make predictions for multiple seasons with no human adaptations. To do
so, various stand-alone models as well as carefully designed complex models
were implemented and evaluated. After repeated experiments, a nal model
that combines deep learning techniques with traditional classi ers was able to
achieve this task.
The nal model uses multi-layer perceptron, a deep learning model, to
extract hidden features within given data, and feed these feature into a gradient
boosting model to make nal predictions. A variety of other techniques were
also used to ensure the reliability and accuracy of the nal model.
Tested for ve seasons, the model outperformed key benchmarks by large
margins. Further, it showed consistent gain of performance over benchmarks.
Its performance has been ranked among the top of leader board in a competi-
tion where contestants develop optimized models for each individual season.
In conclusion, the model showed its capability of capturing hidden patterns
that are general to games across seasons, thus making reliable and informative
predictions across seasons without manual adaptations. |
| first_indexed | 2024-10-01T06:54:08Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/70198 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T06:54:08Z |
| publishDate | 2017 |
| record_format | dspace |
| spelling | ntu-10356/701982023-03-03T20:50:57Z Forecasting sport-matches through data mining II Fu, Jiaxiang Pan Jialin, Sinno School of Computer Science and Engineering Centre for Computational Intelligence DRNTU::Engineering::Computer science and engineering In recent years, machine learning, particularly deep learning has become in- creasingly studied, and applied in different areas of interest. One of them is the prediction of sports game results. Many have attempted to develop models to predict results of National Col- legiate Athletic Association (NCAA) Men's Basketball Tournament. However, existing models require manual e ort to rst extract features from data before they can be trained to make predictions. These manual processes typically need to be repeated when model is applied to new data, a new season of game, for example. This project aims to develop a model that can generalize its training to make predictions for multiple seasons with no human adaptations. To do so, various stand-alone models as well as carefully designed complex models were implemented and evaluated. After repeated experiments, a nal model that combines deep learning techniques with traditional classi ers was able to achieve this task. The nal model uses multi-layer perceptron, a deep learning model, to extract hidden features within given data, and feed these feature into a gradient boosting model to make nal predictions. A variety of other techniques were also used to ensure the reliability and accuracy of the nal model. Tested for ve seasons, the model outperformed key benchmarks by large margins. Further, it showed consistent gain of performance over benchmarks. Its performance has been ranked among the top of leader board in a competi- tion where contestants develop optimized models for each individual season. In conclusion, the model showed its capability of capturing hidden patterns that are general to games across seasons, thus making reliable and informative predictions across seasons without manual adaptations. Bachelor of Engineering (Computer Science) 2017-04-15T07:13:11Z 2017-04-15T07:13:11Z 2017 Final Year Project (FYP) http://hdl.handle.net/10356/70198 en Nanyang Technological University 53 p. application/pdf |
| spellingShingle | DRNTU::Engineering::Computer science and engineering Fu, Jiaxiang Forecasting sport-matches through data mining II |
| title | Forecasting sport-matches through data mining II |
| title_full | Forecasting sport-matches through data mining II |
| title_fullStr | Forecasting sport-matches through data mining II |
| title_full_unstemmed | Forecasting sport-matches through data mining II |
| title_short | Forecasting sport-matches through data mining II |
| title_sort | forecasting sport matches through data mining ii |
| topic | DRNTU::Engineering::Computer science and engineering |
| url | http://hdl.handle.net/10356/70198 |
| work_keys_str_mv | AT fujiaxiang forecastingsportmatchesthroughdataminingii |