Web application for basketball play recommendation and tactic identification
In recent years, the adaptation of digital implementations and computer-powered systems into the sports industry has been ubiquitous. Techniques such as machine learning, image recognition, and big data analysis are being applied to an unprecedented level for various purposes, such as sporting event...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/144624 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In recent years, the adaptation of digital implementations and computer-powered systems into the sports industry has been ubiquitous. Techniques such as machine learning, image recognition, and big data analysis are being applied to an unprecedented level for various purposes, such as sporting event capture, viewership analysis, and recommender optimization. In particular, similar sports play retrieval is one that interests many organizations and individuals.
Similar sports retrieval is the process of fetching plays similar to the query play from a database. The similarity measures are based on certain established or arbitrary metrics. It is often challenging to design and produce a responsive, reliable, and accurate sports play retrieval system due to the unideal speed of retrieval and difficulty in selecting the optimal target plays. Consequently, there exist rather few systems and applications exploring the practical benefits of such a process.
Therefore, by building on a robust and efficient similar play retrieval model called play2vec, this project aims to develop a web application for similar NBA plays recommendation and tactic identification and labelling. The recommended plays and set play labels will be presented to the users with various kinds of visualizations on a web interface.
In order to obtain representative retrievals, basketball tracking data of selected NBA games are collected and preprocessed. The implementations of data processing, model building, and model training were done using Python with open-source libraries. The web application was built using a variety of frameworks and web development tools and was integrated with the retrieval model to function as a recommender based sports application.
This project could be further improved with the availability of tracking data of recent seasons. Moreover, a hosting server machine with higher specifications could be used for faster processing and response speed of the application. |
|---|