Introduction

• 3 Key Questions

  • Can I Predict Fantasy Points Given only Targets as Input?
  • Can I Predict Fantasy Points Given Targets and Yards Per Reception?
  • Can I Predict Touchdowns Given All Inputs?

ESPN Top 3 Week 13 Wide Receiver Projections

Data Overview

Data Overview

Data Overview

Correlation:

Data Overview

Relationship to Fantasy Point Total

Question 1: Fantasy Points Given Targets

• Model 1: Simple Linear Regression

  • FPTS ~ TGT Linear Model

## [1] "R-Squared:  0.792"

Question 1: Fantasy Points Given Targets

• Model 2: XGBoost Linear Regression

  • FPTS ~ TGT Linear Model

Question 1: Fantasy Points Given Targets

• Model 2: XGBoost Linear Regression

##                         XGBoost Model Score
## Mean_Squared_Error                    8.300
## Root_Mean_Squared_Error               2.881
## Test_Mean                             3.141
## R_Squared                             0.728

Comments

The XGBoost Model is prone to overfitting. The parameter, eta, is the main valve to open or close to limit overfitting. A lower eta value will help prevent overfitting.

Question 2: Fantasy Points Given Targets and Yards Per Catch

• Model 3: Multiple Regression

## [1] "R-Squared:  0.827"

Question 2: Fantasy Points Given Targets and Yards Per Catch

• Model 4: Negative Binomial

The Negative Binomial Model provides a very poor fit. With an AIC of 6795.4 and predictions that are immensely far from expectation, this model will not be included.

Question 2: Fantasy Points Given Targets and Yards Per Catch

• Model 5: XGBoost Regression Model with TGT and YPR as Inputs

Question 2: Fantasy Points Given Targets and Yards Per Catch

• Model 5: XGBoost Regression Model with TGT and YPR as Inputs
  • Importance Plot

Question 2: Fantasy Points Given Targets and Yards Per Catch

• Model 5: XGBoost Regression Model with TGT and YPR as Inputs
  • Model Metrics

##                         XGBoost Model Score
## Mean_Squared_Error                    6.217
## Root_Mean_Squared_Error               2.493
## Test_Mean                             2.887
## R_Squared                             0.755

Question 3: Touchdowns Given All Inputs

• Fantasy point totals will not be included in the touchdown prediction

  • Model 6: Multiple Regression
    • As expected, the model falls well short of correctly projecting touchdowns.

## [1] "R-Squared:  0.302"

Question 3: Touchdowns Given All Inputs

• Model 7: XGBoost Touchdowns From All Inputs

Question 3: Touchdowns Given All Inputs

• Model 7: XGBoost Touchdowns From All Inputs
• Importance Plot

Question 3: Touchdowns Given All Inputs

• Model 7: XGBoost Touchdowns From All Inputs
• Model Metrics

##                         XGBoost Model Score
## Mean_Squared_Error                    0.086
## Root_Mean_Squared_Error               0.293
## Test_Mean                             0.233
## R_Squared                            -7.179

Comments

Touchdowns are effectively random. In the knitted document, the seed is different from what it was when the model originally ran, and the \(R^2\) decreased from 0.95 to a negative value.

Model Selection and Week 14 Predictions

Week 14 Projections

Week 14 Performance

Comments

Overall, in Week 14, the models that performed the best were the XGBoost multiple regression and the standard multiple regression model that took into account expected yards per reception and targets.

Sources:

“Fantasy Football Scoring Leaders.” ESPN, ESPN Internet Ventures, https://fantasy.espn.com/football/leaders?lineupSlot=4&scoringPeriodId=13&statSplit=singleScoringPeriod.

“Week 14 NFL WR Statistics.” FantasyPros, https://www.fantasypros.com/nfl/stats/wr.php?range=week&week=14.