Skip to content

mg4432/Kaggle-Battle-ground

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 

Repository files navigation

기계학습을 통한 매치 순위에 영향을 미치는 요인 파악

Kaggle - PUBG Finish Placement Prediction 순위 예측 모델 개발 프로젝트

1. 프로젝트 개요

1.1 Members

1.2 프로젝트 목적

  • 기계학습을 통한 매치 순위에 영향을 미치는 요인 파악
  • 대회 당시(2018)년 리더보드 기준 상위 5% (76/1528) 이상의 성능을 가진 모델 개발

1.3 진행 기간

  • 02/27 ~ 3/25

주차별 계획

  • 1주차(2/27~3/5) : 데이터 내 cheating 유저 파악 & EDA
  • 2주차(3/6~3/12) : EDA 기반 Feature Engineering & Baseline 모델 생성, 세부적 분석 방법 조정
  • 3주차(3/13~3/18) : 분석 계획에 따른 분석 모델 선정, 생성, 학습 및 고도화
  • 4주차(3/19~3/25) : 분석 결과, 분석 방법, 분석 상의 한계점 및 개선사항 정리 후 보고서 작성

image

1.4 Contests

  1. Project description

Project outline, purpose, weekly plan, etc.

  1. Project process

Hypotheses, EDA, feature engineering, model selection, etc.

  1. Our strategy

Stretegies used to develop performance and save time & memory, etc.

  1. Result & Conclustion

Our rank in LB, MAE Most effective factors in predicting finish placement.

2. Process

2.1 Overall process

Baseline → Pre-processing(& Feature engineering) → Group place → Post-processing

image

2.2 Hypotheses

2.2.1 Hypotheses

image

2.2.2 Results

image

image

image

2.3 Model selection

image

3. Our strategy

3.1 Reduce memory

  • 메모리, 시간을 절약하기 위해 데이터 타입 변경

image

3.2 Get_group_place :

  • killPlace 변수를 통해서 그룹 간의 순위 관계를 추론함.

image

3.2.1 Case 1 : Orders can be inferred exactly

image

3.2.2 Case 2 : Orders cannot be inferred exactly

image

image

  • Post-processing : image

4. 프로젝트 결과

4.1 매치 순위에 영향을 주는 요인

image

높은 순위를 차지하기 위해서 유저들이 취하는 방법에는 크게 2가지가 있음. 

1. 교전을 통해서 적을 제거하는 방법 
2. 교전을 피하면서 오래 살아남는 방법 

분석 결과, 높은 등수를 차지한 그룹은 교전을 피하는 그룹이 아니라 전투력이 높고 교전을 많이 진행한 그룹이었음을 확인할 수 있었음.
따라서, 매치 순위에 영향을 주는 요인은 그룹의 전투력이라고 할 수 있음.

4.2 Rank & MAE

Rank : 5 / 1528 (0.327%) 
MAE : 0.01836

image

5. File directory

📂 Project
├── 📂 code
│    ├── Baseline.ipynb   # Baseline model
│    ├── EDA.ipynb   
│    ├── get_group_place_function.ipynb   # Function for getting group places
│    ├── postprocessing.ipynb   # Postprocessing predicted valuefrom continuous to discrete
│    └── LightGBM_final.ipynb   # Final model
├── 📂 dataset
│    ├── train_sample.csv   # train data sample
│    ├── test_sample.csv   # test data sample
│    └── sampleSubmission_sample.csv   # submission file sample
└── 📂 final
     ├── report.pptx 
     ├── report.pdf 
     └── submission.csv # final submission file

About

Kaggle - PUBG Finish Placement Prediction 순위 예측 모델 개발 프로젝트

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published