πŸ¦†
Statduck
  • (The link was changed)
  • Resume
  • Projects
    • KBO Prediction
    • Building's Fire in Gimhae
    • Parking Demand
    • Profiling of Y&Z Investors
    • Atlanta Police Station
    • Meal kit location strategy
  • Machine Learning
    • Overview: Supervised Learning
      • Curse of dimensionality
      • Statistical Decision Theory
    • Linear method: Regression
      • Problem Definition
      • Residual Sum of Squares
      • Coefficients(Beta)
      • Orthogonalization
      • Variation Version
      • Some problems
    • Linear method: Classification
      • Linear Model
      • LDA & QDA
      • Generalized Discriminant Analysis
      • Calculation
      • RDA
      • Support Vector Machine
    • Basis Expansions & Regularization
      • Piecewise Polynomials and Splines
      • Natural Cubic Spline
      • Smoothing Splines
      • Smoother Matrices
      • λΉ„λͺ¨μˆ˜ λ‘œμ§€μŠ€ν‹± νšŒκ·€
    • Kernel Smoothing
      • Kernel method
      • Local regression
      • Kernel Density Estimation
      • Local Likelihood
    • Optimization
      • Convex Optimization
      • Conditional Gradient
      • Novelty Detection
    • Model Assessment & Selection
      • Bias-Var decomposition
      • Optimism
      • The Effective Number of Parameters
      • CV and Boostrap
    • Ensemble
      • Boosting
    • Unsupervised Learning
      • Association Rule
      • Clustering & Nearest Neighbor
      • Self-Organized Map
      • K-medoids[PAM algorithm]
  • Bayes
    • Prior, Posterior, Sample
    • One Parameter Model
    • Two parameter model
    • Bayesian Regression
  • Python Study
    • Python OverView
    • Class
    • Algorithm
      • Algorithm Analysis
      • Recursion
      • Binary Search
      • Stack
      • DFS
  • Mathematical Stat
    • Probability distribution
    • Discrete Variable
    • Moment Generating Function
  • Recommender System
    • LightGCN
    • Collaborative Filtering
      • Neighborhood-Based method
      • A View of Regression
      • Graph Model for NB model
    • Model-Based CF
      • Latent Factor Models
      • Subproblem
      • Summary
    • Content based filtering
    • Ensemble-Based
  • Incomplete posting
    • D3-js[Data handling]
    • Ch6(editing)
    • Gibbs Sampling
    • Session 9(editing)
  • Causal inference
  • SQL
    • Introduction
    • Basic Statement
    • Join Statement
Powered by GitBook
On this page

Was this helpful?

  1. Projects

KBO Prediction

2020 Predicting Winning Rates, AVG, and ERA by team in KBO

PreviousProjectsNextBuilding's Fire in Gimhae

Last updated 2 years ago

Was this helpful?

The competition is held by Big-contest

  • Problem: 2020 Prediction of winning rates, avg, and era by team in KBO(Korea Baseball Organization) regular season. The several records until 2020.07 are used to predict several indicators during new season starting from 2020.09.

  • Challenge: We don't have any record in future(2020.09), so prediction of XXX values is required.

  • Approach: Because the duplicated prediction with X^\hat{X}X^=>Y^\hat{Y}Y^ shows a high error, so we just predict Y^\hat{Y}Y^ directly without X^\hat{X}X^. We used univariate time series forecasting model. However, it was hard to find the linear relation between past data and present data, so that we adjust our ARIMA model to fit a non-linear structure by using random forest. This model captures the behavior occurred in the past. Baseball has a tendency in winning, we also make a regression model on data by year(tendency of whole year) and by 10 days(tendency of recent plays). By averaging theses three results, we predicted the indicators for baseball game,

  • Results: Our model shows the team NC has 0.25 winning rate even it had the highest winning rate 0.628 before 20.09.28. In this period this Team really shows low winning rate: 0.428. Our team advanced to final in big contest.

  • Evaluation: We could't explain the reason why our model works clearly. To work out this issue, the deeper understanding of time-series analysis was needed.

  • Period: 2020.07-2020.11

  • Prize: Advanced to the finals