Inspired by Kaggle’s 2018 NCAA March Madness Contest, I created a prediction model for the results of the 2018 NCAA Men’s Basketball tournament against the spread and am keeping track of the results. This document will keep a running track of all the predictions and the model’s accuracy.
There were seven tasks to be repeated in each round:
The seven functions below accomplish those tasks.
# Predict outcomes and input into bracket
predict_ncaa_round <- function(round){
# Subset data by round
if(round == 0){rows <- c(1:4)}
if(round == 1){rows <- c(5:36)}
if(round == 2){rows <- c(37:52)}
if(round == 3){rows <- c(53:60)}
if(round == 4){rows <- c(61:64)}
if(round == 5){rows <- c(65:66)}
if(round == 6){rows <- 67}
# Append model predicted spread for higher seeds
bracket <- bracket[rows,] %>%
left_join(select(PredictionData, TeamName.x, TeamName.y, Prediction, Win_Prob), by = c('TeamName.x' = 'TeamName.x', 'TeamName.y'= 'TeamName.y'))
# Get the rows of lesser seeds
na_rows <- which(is.na(bracket$Prediction))
# Impute prediction data from lesser seeds
for (i in na_rows) {
bracket[i, 7] <- -PredictionData[which(bracket[i, 5] == PredictionData$TeamName.y & bracket[i,6] == PredictionData$TeamName.x), 9]
bracket[i, 8] <- 1 - PredictionData[which(bracket[i, 5] == PredictionData$TeamName.y & bracket[i,6] == PredictionData$TeamName.x), 10]
}
bracket
}# Appending Sportsbook Spreads
attachSpreads <- function(round, spreadDF){
DF <- spreadDF
# Append actual spreads from bookies
names(DF) <- c("Team", "Spread")
DF$Spread <- as.numeric(DF$Spread)
# Append spreads on to bracket
round <- round %>%
left_join(DF, by = c('TeamName.x' = 'Team')) %>%
mutate(Diff = Prediction - Spread)
round
}# Appending Game Results
attachGameResults <- function(round, resultsDF){
DF <- resultsDF
names(DF) <- c("Team", "Team.x.score", "Team.y.score")
DF$Team.x.score <- as.numeric(DF$Team.x.score)
DF$Team.y.score <- as.numeric(DF$Team.y.score)
# Append Game Results to Bracket
round <- round %>%
left_join(DF, by = c('TeamName.x' = 'Team'))
round
}# Determine who model would bet on ATS
ATSchoice <- function(bracket){
bracket$ModelChoice <- NA
for (i in c(1:length(bracket$Diff))){
if(bracket$Diff[i] < -2)
{bracket$ModelChoice[i] <- bracket$TeamName.y[i]}
if(bracket$Diff[i] > 2)
{bracket$ModelChoice[i] <- bracket$TeamName.x[i]}
if(bracket$Diff[i] <= 2 & bracket$Diff[i] >= -2)
{bracket$ModelChoice[i] <- "No choice"}
}
bracket
}# Assessing Model Accuracy ATS
ATSresults <- function(bracket){
# Determine ATS winner
bracket$ATSWinner <- NA
for (i in c(1:length(bracket$Diff))){
FinalDiff <- bracket$Team.x.score[i] - bracket$Team.y.score[i]
if(!is.na(FinalDiff)) {
if(FinalDiff < bracket$Spread[i])
{bracket$ATSWinner[i] <- bracket$TeamName.y[i]}
if(FinalDiff > bracket$Spread[i])
{bracket$ATSWinner[i] <- bracket$TeamName.x[i]}
if(FinalDiff == bracket$Spread[i])
{bracket$ATSWinner[i] <- "Push"}
}
}
# Determine if Model Prediction was Correct, Incorrect, or Not Made
bracket$Accuracy <- NA
for (i in c(1:length(bracket$Diff))){
if(!is.na(bracket$ATSWinner[i])) {
if(bracket$ModelChoice[i] == bracket$ATSWinner[i])
{bracket$Accuracy[i] <- "Correct"}
if(bracket$ModelChoice[i] != bracket$ATSWinner[i])
{bracket$Accuracy[i] <- "Incorrect"}
if(bracket$ATSWinner[i] == "Push")
{bracket$Accuracy[i] <- "Push"}
}
if(is.na(bracket$ATSWinner[i]))
{bracket$Accuracy[i] <- NA}
if(bracket$ModelChoice[i] == "No choice")
{bracket$Accuracy[i] <- NA}
}
bracket
}# Putting winners into next round
advance_winners <- function(x){
for (i in c(1:length(x$Slot))){
slot <- x$Slot[i]
if(x$Team.x.score[i] > x$Team.y.score[i]) {
winner <- x$TeamName.x[i]
} else{
winner <- x$TeamName.y[i]
}
rownum <- which(bracket[,c(3:4)] == slot, arr.ind = TRUE)[1]
colnum <- which(bracket[,c(3:4)] == slot, arr.ind = TRUE)[2]
if(colnum == 1)
{bracket$TeamName.x[rownum] <- winner}
else
{bracket$TeamName.y[rownum] <- winner}
}
bracket
}# Create Prediction Table
predictionTable <- function(bracket){
predictionstable <- bracket %>%
select(StrongSeed, WeakSeed, TeamName.x, TeamName.y, Prediction, Spread, ModelChoice, Diff) %>%
arrange(desc(abs(Diff)))
predictionstable %>%
kable("html") %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed"), font_size = 9) %>%
footnote(general = "A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points.")
}
# Create Results & Accuracy Table
resultsTable <- function(bracket){
results_table <- bracket %>%
select(StrongSeed, TeamName.x, WeakSeed, TeamName.y, Prediction, Spread, ModelChoice, Team.x.score, Team.y.score, ATSWinner, Accuracy)
results_table %>%
kable("html") %>%
kable_styling(bootstrap_options = c("striped", "hover", "condensed"), font_size = 9)
}Throughout the tournament, spreads were gathered manually from the internet and put into these tables. The date and time of the spreads are recorded. Spreads are positive if strong seed is favored, negative if underdogs.
# Playin Spreads retroactively added after the games
playinspreads <- data.frame(
matrix(
c(
"St Bonaventure", -3.5,
"Long Island", -4.5,
"Arizona St", 1,
"NC Central", -4.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# Round 1 Spreads from Wednesday night
round1spreads <- data.frame(
matrix(
c(
"Rhode Island", 1.5,
"Tennessee", 11.5,
"Gonzaga", 13.5,
"Kansas", 14.5,
"Duke", 20,
"Miami FL", 2,
"Ohio St", 8,
"Seton Hall", 3,
"Villanova", 23,
"Kentucky", 5,
"Houston", 4,
"Texas Tech", 11,
"Virginia Tech", 2.5,
"Arizona", 8.5,
"Michigan", 10,
"Florida", 5.5,
"Texas A&M", 2.5,
"Purdue", 20.5,
"Wichita St", 11.5,
"Cincinnati", 14,
"North Carolina", 19.5,
"Arkansas", -1.5,
"West Virginia", 10.5,
"Nevada", 0,
"Creighton", 1,
"Michigan St", 14.5,
"Xavier", 19.5,
"Auburn", 9,
"Virginia", 21,
"TCU", 4,
"Missouri", -1.5,
"Clemson", 4.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# Round 2 Spreads from Saturday 11:30am
round2spreads <- data.frame(
matrix(
c(
"Villanova", 11,
"Purdue", 3.5,
"Texas Tech", 0,
"Marshall", -11,
"Kansas", 4.5,
"Duke", 10,
"Michigan St", 9,
"Auburn", 1,
"UMBC", -12,
"Cincinnati", 7,
"Tennessee", 6,
"Buffalo", -7.5,
"Xavier", 4,
"North Carolina", 6,
"Michigan", 3,
"Gonzaga", 2.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# Round 3 (Sweet 16) Spreads from 7:00pm Monday
round3spreads <- data.frame(
matrix(
c(
"Villanova", 5.5,
"Purdue", 1.5,
"Kansas", 4.5,
"Duke", 11,
"Kansas St", -6,
"Nevada", 1.5,
"Florida St", -5.5,
"Texas A&M", -2.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# All Round 4 Spreads from 7:00pm Monday
round4spreads <- data.frame(
matrix(
c(
"Villanova", 7,
"Kansas", -3.5,
"Kansas St", 1.5,
"Florida St", -4.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# All Round 5 Spreads from 8:30pm Thursday
round5spreads <- data.frame(
matrix(
c(
"Villanova", 5,
"Loyola-Chicago", -5.5),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)
# All Round 6 Spreads from Sunday at 10:30pm
round6spreads <- data.frame(
matrix(
c(
"Villanova", 7),
ncol = 2, byrow = TRUE), stringsAsFactors = FALSE)Throughout the tournament, the game results were gathered manually from the internet and put into these tables. This task needs to eventually be automated.
# Playin Game Results
playinresults <- data.frame(
matrix(
c(
"St Bonaventure", 65, 58,
"Long Island", 61, 71,
"Arizona St", 56, 60,
"NC Central", 46, 64), ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 1 Game Results
round1results <- data.frame(
matrix(
c(
"St Bonaventure", 65, 58,
"Long Island", 61, 71,
"Arizona St", 56, 60,
"NC Central", 46, 64,
"Rhode Island", 83, 78,
"Tennessee", 73, 47,
"Gonzaga", 68, 64,
"Kansas", 76, 60,
"Duke", 89, 67,
"Miami FL", 62, 64,
"Ohio St", 81, 73,
"Seton Hall", 94, 83,
"Villanova", 87, 61,
"Kentucky", 78, 73,
"Houston", 67, 65,
"Texas Tech", 70, 60,
"Virginia Tech", 83, 86,
"Arizona", 68, 89,
"Michigan", 61, 47,
"Florida", 77, 62,
"Texas A&M", 73, 69,
"Purdue", 74, 48,
"Wichita St", 75, 81,
"Cincinnati", 68, 53,
"North Carolina", 84, 66,
"Arkansas", 62, 79,
"West Virginia", 85, 68,
"Nevada", 87, 83,
"Creighton", 59, 69,
"Michigan St", 82, 78,
"Xavier", 91, 72,
"Auburn", 62, 58,
"Virginia", 54, 74,
"TCU", 52, 57,
"Missouri", 54, 67,
"Clemson", 79, 68), ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 2 Game Results
round2results <- data.frame(
matrix(
c(
"Villanova", 81, 58,
"Texas Tech", 69, 66,
"Duke", 87, 62,
"Tennessee", 62, 63,
"Buffalo", 75, 95,
"Michigan", 64, 63,
"Gonzaga", 90, 84,
"Kansas", 83, 79,
"Purdue", 76, 73,
"Michigan St", 53, 55,
"North Carolina", 65, 86,
"Cincinnati", 73, 75,
"Auburn", 53, 84,
"UMBC", 43, 50,
"Xavier", 70, 75,
"Marshall", 71, 94),
ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 3 Game Results
round3results <- data.frame(
matrix(
c(
"Villanova", 90, 78,
"Purdue", 65, 78,
"Kansas", 80, 76,
"Duke", 69, 65,
"Kansas St", 61, 58,
"Nevada", 68, 69,
"Florida St", 75, 60,
"Texas A&M", 72, 99),
ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 4 Game Results
round4results <- data.frame(
matrix(
c(
"Villanova", 71, 59,
"Kansas", 85, 81,
"Kansas St", 62, 78,
"Florida St", 54, 58),
ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 5 Game Results
round5results <- data.frame(
matrix(
c(
"Villanova", 95, 79,
"Loyola-Chicago", 57, 69),
ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)
# Round 6 Game Results
round6results <- data.frame(
matrix(
c(
"Villanova", 79, 62),
ncol = 3, byrow = TRUE), stringsAsFactors = FALSE)The prediction model is an XGBoost model with 5-fold cross validation. The cleaning and preparation of the data and the training and testing of that model are in a script which can be found here [not releasing at the moment]. The prediction data for all possible pairs of opponents are saved into the file “saferesults.Rds”. The prediction data is not Bayesian; that is, the model predictions are strictly based on pre-tournament performance.
# Load prediction data from model
PredictionData <- readRDS("saferesults.Rds")# Load the 2018 bracket data and team names
tourneyslots <- fread("NCAATourneySlots.csv") %>% filter(Season == 2018)
tourneySeeds <- fread('NCAATourneySeeds.csv') %>% filter(Season == 2018)
teams <- fread('teams.csv')
# Attach team names to seed info
tourneySeeds <- tourneySeeds %>%
left_join(teams, by = 'TeamID') %>%
select(Season, Seed, TeamID, TeamName)
# Manually set the play-in winners
tourneyslots[5, 4] <- "W16b"
tourneyslots[10, 4] <- "W11a"
tourneyslots[18, 4] <- "X11b"
tourneyslots[29, 4] <- "Z16b"
# Complete Bracket
bracket <- tourneyslots %>%
left_join(select(tourneySeeds, -c(Season, TeamID)), by=c('StrongSeed'='Seed')) %>%
left_join(select(tourneySeeds, -c(Season, TeamID)), by=c('WeakSeed'='Seed')) First, we will append the model predicted spreads, the bookie spreads, and the differences between the two on to the bracket.
# Append model predicted spreads to bracket
playin <- predict_ncaa_round(round = 0)
# Append actual spreads from sportsbook
playin <- attachSpreads(playin, playinspreads)A team needs to have a predicted spread differing by more than 2 points from the sportsbook spread in order to have greater than a 55% chance of winning ATS (against the spread), which is the minimum winning percent needed to break even. In the ModelChoice variable, the model will return “No Choice” if the model prediction is within 2 points of the sportsbook spread or otherwise return which team the model predicts has a > 55% chance to cover the spread.
# Determine who model would bet on ATS
playin <- ATSchoice(playin)
# Output of Model Predictions ATS
predictionTable(playin)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| Z16a | Z16b | NC Central | TX Southern | 9.3 | -4.5 | NC Central | 13.8 |
| W11a | W11b | St Bonaventure | UCLA | 2.6 | -3.5 | St Bonaventure | 6.1 |
| W16a | W16b | Long Island | Radford | 0.9 | -4.5 | Long Island | 5.4 |
| X11a | X11b | Arizona St | Syracuse | 1.3 | 1.0 | No choice | 0.3 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The model recommends: NC Central, St Bonaventure, and Long Island.
# Append Final Scores of Games
playin <- attachGameResults(playin, playinresults)
# Determine ATS winner and Model's Accuracy ATS
playin <- ATSresults(playin)
# Table of all outcomes
resultsTable(playin)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| W11a | St Bonaventure | W11b | UCLA | 2.6 | -3.5 | St Bonaventure | 65 | 58 | St Bonaventure | Correct |
| W16a | Long Island | W16b | Radford | 0.9 | -4.5 | Long Island | 61 | 71 | Radford | Incorrect |
| X11a | Arizona St | X11b | Syracuse | 1.3 | 1.0 | No choice | 56 | 60 | Syracuse | NA |
| Z16a | NC Central | Z16b | TX Southern | 9.3 | -4.5 | NC Central | 46 | 64 | TX Southern | Incorrect |
# Summary of results
table(playin$Accuracy)##
## Correct Incorrect
## 1 2The model’s record was 1-2 ATS for the play-in games.
# Append model predicted spreads to bracket
round1 <- predict_ncaa_round(1)
# Append actual spreads from sportsbook
round1 <- attachSpreads(round1, round1spreads)
# Determine who model would bet on ATS
round1 <- ATSchoice(round1)As in the playin round, the model will return “No Choice” if the model prediction is within 2 points of the sportsbook spread. Otherwise, the model returned which team it predicts has a > 55% chance to cover the spread.
There is something wrong with my predictions for the top seeds, especially 1s and 2s, so I stayed away from following my model on any of those. Other kagglers have had this same issue and have manually input values for them, some trouble I didn’t want to bother with until I had the rest of the model running well. I’ll focus on that at some point in the future. For now, I’m going to manually input “No Choice” for all 1 and 2 seeds of Villanova, Kansas, Duke, Purdue, Cincinnati, Xavier, UNC, and Virginia.
# Manually set 1 and 2 seeds to no choice
round1$ModelChoice[c(1,2,9,10,17,18,25,26)] <- "No choice"# Output of Model Predictions ATS
predictionTable(round1)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| Y01 | Y16 | Virginia | UMBC | 3.5 | 21.0 | No choice | -17.5 |
| Y04 | Y13 | Arizona | Buffalo | -7.1 | 8.5 | Buffalo | -15.6 |
| W01 | W16b | Villanova | Radford | 9.5 | 23.0 | No choice | -13.5 |
| Z08 | Z09 | Missouri | Florida St | -11.0 | -1.5 | Florida St | -9.5 |
| W05 | W12 | West Virginia | Murray St | 1.1 | 10.5 | Murray St | -9.4 |
| X04 | X13 | Auburn | Col Charleston | -0.1 | 9.0 | Col Charleston | -9.1 |
| X01 | X16 | Kansas | Penn | 6.8 | 14.5 | No choice | -7.7 |
| W04 | W13 | Wichita St | Marshall | 4.3 | 11.5 | Marshall | -7.2 |
| W02 | W15 | Purdue | CS Fullerton | 13.4 | 20.5 | No choice | -7.1 |
| Y02 | Y15 | Cincinnati | Georgia St | 7.1 | 14.0 | No choice | -6.9 |
| X03 | X14 | Michigan St | Bucknell | 7.8 | 14.5 | Bucknell | -6.7 |
| W08 | W09 | Virginia Tech | Alabama | 9.2 | 2.5 | Virginia Tech | 6.7 |
| Z02 | Z15 | North Carolina | Lipscomb | 13.1 | 19.5 | No choice | -6.4 |
| Y07 | Y10 | Nevada | Texas | 6.2 | 0.0 | Nevada | 6.2 |
| Y03 | Y14 | Tennessee | Wright St | 5.4 | 11.5 | Wright St | -6.1 |
| Z04 | Z13 | Gonzaga | UNC Greensboro | 7.8 | 13.5 | UNC Greensboro | -5.7 |
| X02 | X15 | Duke | Iona | 15.0 | 20.0 | No choice | -5.0 |
| X05 | X12 | Clemson | New Mexico St | -0.3 | 4.5 | New Mexico St | -4.8 |
| Z03 | Z14 | Michigan | Montana | 5.2 | 10.0 | Montana | -4.8 |
| Y06 | Y11 | Miami FL | Loyola-Chicago | 6.7 | 2.0 | Miami FL | 4.7 |
| Z01 | Z16b | Xavier | TX Southern | 15.6 | 19.5 | No choice | -3.9 |
| Y05 | Y12 | Kentucky | Davidson | 1.7 | 5.0 | Davidson | -3.3 |
| X07 | X10 | Rhode Island | Oklahoma | 4.7 | 1.5 | Rhode Island | 3.2 |
| X06 | X11b | TCU | Syracuse | 1.0 | 4.0 | Syracuse | -3.0 |
| W06 | W11a | Florida | St Bonaventure | 8.0 | 5.5 | Florida | 2.5 |
| Y08 | Y09 | Creighton | Kansas St | -1.1 | 1.0 | Kansas St | -2.1 |
| X08 | X09 | Seton Hall | NC State | 1.1 | 3.0 | No choice | -1.9 |
| Z06 | Z11 | Houston | San Diego St | 5.1 | 4.0 | No choice | 1.1 |
| Z05 | Z12 | Ohio St | S Dakota St | 8.9 | 8.0 | No choice | 0.9 |
| W03 | W14 | Texas Tech | SF Austin | 11.4 | 11.0 | No choice | 0.4 |
| Z07 | Z10 | Texas A&M | Providence | 2.3 | 2.5 | No choice | -0.2 |
| W07 | W10 | Arkansas | Butler | -1.4 | -1.5 | No choice | 0.1 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
# Append Final Scores of Games
round1 <- attachGameResults(round1, round1results)
# Determine ATS winner and Model's Accuracy ATS
round1 <- ATSresults(round1)
# Table of Outcomes in Round 1
resultsTable(round1)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| W01 | Villanova | W16b | Radford | 9.5 | 23.0 | No choice | 87 | 61 | Villanova | NA |
| W02 | Purdue | W15 | CS Fullerton | 13.4 | 20.5 | No choice | 74 | 48 | Purdue | NA |
| W03 | Texas Tech | W14 | SF Austin | 11.4 | 11.0 | No choice | 70 | 60 | SF Austin | NA |
| W04 | Wichita St | W13 | Marshall | 4.3 | 11.5 | Marshall | 75 | 81 | Marshall | Correct |
| W05 | West Virginia | W12 | Murray St | 1.1 | 10.5 | Murray St | 85 | 68 | West Virginia | Incorrect |
| W06 | Florida | W11a | St Bonaventure | 8.0 | 5.5 | Florida | 77 | 62 | Florida | Correct |
| W07 | Arkansas | W10 | Butler | -1.4 | -1.5 | No choice | 62 | 79 | Butler | NA |
| W08 | Virginia Tech | W09 | Alabama | 9.2 | 2.5 | Virginia Tech | 83 | 86 | Alabama | Incorrect |
| X01 | Kansas | X16 | Penn | 6.8 | 14.5 | No choice | 76 | 60 | Kansas | NA |
| X02 | Duke | X15 | Iona | 15.0 | 20.0 | No choice | 89 | 67 | Duke | NA |
| X03 | Michigan St | X14 | Bucknell | 7.8 | 14.5 | Bucknell | 82 | 78 | Bucknell | Correct |
| X04 | Auburn | X13 | Col Charleston | -0.1 | 9.0 | Col Charleston | 62 | 58 | Col Charleston | Correct |
| X05 | Clemson | X12 | New Mexico St | -0.3 | 4.5 | New Mexico St | 79 | 68 | Clemson | Incorrect |
| X06 | TCU | X11b | Syracuse | 1.0 | 4.0 | Syracuse | 52 | 57 | Syracuse | Correct |
| X07 | Rhode Island | X10 | Oklahoma | 4.7 | 1.5 | Rhode Island | 83 | 78 | Rhode Island | Correct |
| X08 | Seton Hall | X09 | NC State | 1.1 | 3.0 | No choice | 94 | 83 | Seton Hall | NA |
| Y01 | Virginia | Y16 | UMBC | 3.5 | 21.0 | No choice | 54 | 74 | UMBC | NA |
| Y02 | Cincinnati | Y15 | Georgia St | 7.1 | 14.0 | No choice | 68 | 53 | Cincinnati | NA |
| Y03 | Tennessee | Y14 | Wright St | 5.4 | 11.5 | Wright St | 73 | 47 | Tennessee | Incorrect |
| Y04 | Arizona | Y13 | Buffalo | -7.1 | 8.5 | Buffalo | 68 | 89 | Buffalo | Correct |
| Y05 | Kentucky | Y12 | Davidson | 1.7 | 5.0 | Davidson | 78 | 73 | Push | Push |
| Y06 | Miami FL | Y11 | Loyola-Chicago | 6.7 | 2.0 | Miami FL | 62 | 64 | Loyola-Chicago | Incorrect |
| Y07 | Nevada | Y10 | Texas | 6.2 | 0.0 | Nevada | 87 | 83 | Nevada | Correct |
| Y08 | Creighton | Y09 | Kansas St | -1.1 | 1.0 | Kansas St | 59 | 69 | Kansas St | Correct |
| Z01 | Xavier | Z16b | TX Southern | 15.6 | 19.5 | No choice | 91 | 72 | TX Southern | NA |
| Z02 | North Carolina | Z15 | Lipscomb | 13.1 | 19.5 | No choice | 84 | 66 | Lipscomb | NA |
| Z03 | Michigan | Z14 | Montana | 5.2 | 10.0 | Montana | 61 | 47 | Michigan | Incorrect |
| Z04 | Gonzaga | Z13 | UNC Greensboro | 7.8 | 13.5 | UNC Greensboro | 68 | 64 | UNC Greensboro | Correct |
| Z05 | Ohio St | Z12 | S Dakota St | 8.9 | 8.0 | No choice | 81 | 73 | Push | NA |
| Z06 | Houston | Z11 | San Diego St | 5.1 | 4.0 | No choice | 67 | 65 | San Diego St | NA |
| Z07 | Texas A&M | Z10 | Providence | 2.3 | 2.5 | No choice | 73 | 69 | Texas A&M | NA |
| Z08 | Missouri | Z09 | Florida St | -11.0 | -1.5 | Florida St | 54 | 67 | Florida St | Correct |
# Summary of all tourney results
entiretourney <- rbind(playin, round1)
table(entiretourney$Accuracy)##
## Correct Incorrect Push
## 12 8 1The model’s record was 11-6-1 ATS in round 1 action and is 12-8-1 ATS over the entire tournament. The model would have been 3-5 in the 1 and 2 seeded games.
One last interesting note is that a lot of 1 seeds seemed to not cover the first half spread. I heard this has some historical traction to it. Something to keep in mind next year.
# Recording Round 1 Winners into Round 2
bracket <- advance_winners(round1)The model predictions for all round 2 games are on the rightmost column in the output table below.
# Append model predicted spreads to bracket
round2 <- predict_ncaa_round(2)
# Append actual spreads from sportsbook
round2 <- attachSpreads(round2, round2spreads)
# Determine who model would bet on ATS
round2 <- ATSchoice(round2)
# Output of model recommendations
predictionTable(round2)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| R1Y4 | R1Y5 | Buffalo | Kentucky | 2.2 | -7.5 | Buffalo | 9.7 |
| R1Y1 | R1Y8 | UMBC | Kansas St | -3.5 | -12.0 | UMBC | 8.5 |
| R1Z4 | R1Z5 | Gonzaga | Ohio St | 10.6 | 2.5 | Gonzaga | 8.1 |
| R1W4 | R1W5 | Marshall | West Virginia | -4.6 | -11.0 | Marshall | 6.4 |
| R1X2 | R1X7 | Duke | Rhode Island | 3.8 | 10.0 | Rhode Island | -6.2 |
| R1W2 | R1W7 | Purdue | Butler | 8.1 | 3.5 | Purdue | 4.6 |
| R1Z1 | R1Z8 | Xavier | Florida St | -0.2 | 4.0 | Florida St | -4.2 |
| R1X3 | R1X6 | Michigan St | Syracuse | 5.2 | 9.0 | Syracuse | -3.8 |
| R1Z2 | R1Z7 | North Carolina | Texas A&M | 3.3 | 6.0 | Texas A&M | -2.7 |
| R1W1 | R1W8 | Villanova | Alabama | 8.3 | 11.0 | Alabama | -2.7 |
| R1X1 | R1X8 | Kansas | Seton Hall | 2.2 | 4.5 | Seton Hall | -2.3 |
| R1Y2 | R1Y7 | Cincinnati | Nevada | 9.1 | 7.0 | Cincinnati | 2.1 |
| R1X4 | R1X5 | Auburn | Clemson | -0.6 | 1.0 | No choice | -1.6 |
| R1W3 | R1W6 | Texas Tech | Florida | -0.7 | 0.0 | No choice | -0.7 |
| R1Z3 | R1Z6 | Michigan | Houston | 3.5 | 3.0 | No choice | 0.5 |
| R1Y3 | R1Y6 | Tennessee | Loyola-Chicago | 5.8 | 6.0 | No choice | -0.2 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The model is still picking against all #1 seeds, but no longer against all #2 seeds. So I’m going to start evaluating model on all recommendations, while still hesitant to wager on #1 seed games. The recommendations are:
# Append Final Scores of Games
round2 <- attachGameResults(round2, round2results)
# Determine ATS winner and Model's Accuracy ATS
round2 <- ATSresults(round2)
# Table of round 2 results and outcomes
resultsTable(round2)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| R1W1 | Villanova | R1W8 | Alabama | 8.3 | 11.0 | Alabama | 81 | 58 | Villanova | Incorrect |
| R1W2 | Purdue | R1W7 | Butler | 8.1 | 3.5 | Purdue | 76 | 73 | Butler | Incorrect |
| R1W3 | Texas Tech | R1W6 | Florida | -0.7 | 0.0 | No choice | 69 | 66 | Texas Tech | NA |
| R1W4 | Marshall | R1W5 | West Virginia | -4.6 | -11.0 | Marshall | 71 | 94 | West Virginia | Incorrect |
| R1X1 | Kansas | R1X8 | Seton Hall | 2.2 | 4.5 | Seton Hall | 83 | 79 | Seton Hall | Correct |
| R1X2 | Duke | R1X7 | Rhode Island | 3.8 | 10.0 | Rhode Island | 87 | 62 | Duke | Incorrect |
| R1X3 | Michigan St | R1X6 | Syracuse | 5.2 | 9.0 | Syracuse | 53 | 55 | Syracuse | Correct |
| R1X4 | Auburn | R1X5 | Clemson | -0.6 | 1.0 | No choice | 53 | 84 | Clemson | NA |
| R1Y1 | UMBC | R1Y8 | Kansas St | -3.5 | -12.0 | UMBC | 43 | 50 | UMBC | Correct |
| R1Y2 | Cincinnati | R1Y7 | Nevada | 9.1 | 7.0 | Cincinnati | 73 | 75 | Nevada | Incorrect |
| R1Y3 | Tennessee | R1Y6 | Loyola-Chicago | 5.8 | 6.0 | No choice | 62 | 63 | Loyola-Chicago | NA |
| R1Y4 | Buffalo | R1Y5 | Kentucky | 2.2 | -7.5 | Buffalo | 75 | 95 | Kentucky | Incorrect |
| R1Z1 | Xavier | R1Z8 | Florida St | -0.2 | 4.0 | Florida St | 70 | 75 | Florida St | Correct |
| R1Z2 | North Carolina | R1Z7 | Texas A&M | 3.3 | 6.0 | Texas A&M | 65 | 86 | Texas A&M | Correct |
| R1Z3 | Michigan | R1Z6 | Houston | 3.5 | 3.0 | No choice | 64 | 63 | Houston | NA |
| R1Z4 | Gonzaga | R1Z5 | Ohio St | 10.6 | 2.5 | Gonzaga | 90 | 84 | Gonzaga | Correct |
# Summary of all tourney results
entiretourney <- rbind(entiretourney, round2)
table(entiretourney$Accuracy)##
## Correct Incorrect Push
## 18 14 1In round 2, the model was 6-6 ATS. In total, the model is 18-14-1 through the first two rounds of the tournament for an overall success rate of 56.25%.
# Recording Round 2 Winners into Round 3
bracket <- advance_winners(round2)The model predictions for all round 3 games are on the rightmost column in the output table below.
# Append model predicted spreads to bracket
round3 <- predict_ncaa_round(3)
# Append actual spreads from sportsbook
round3 <- attachSpreads(round3, round3spreads)
# Determine who model would bet on ATS
round3 <- ATSchoice(round3)
# Output of model recommendations
predictionTable(round3)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| R2X2 | R2X3 | Duke | Syracuse | 4.3 | 11.0 | Syracuse | -6.7 |
| R2Y1 | R2Y4 | Kansas St | Kentucky | -2.3 | -6.0 | Kansas St | 3.7 |
| R2X1 | R2X4 | Kansas | Clemson | 8.2 | 4.5 | Kansas | 3.7 |
| R2W1 | R2W4 | Villanova | West Virginia | 2.9 | 5.5 | West Virginia | -2.6 |
| R2Y2 | R2Y3 | Nevada | Loyola-Chicago | -0.8 | 1.5 | Loyola-Chicago | -2.3 |
| R2W2 | R2W3 | Purdue | Texas Tech | 0.6 | 1.5 | No choice | -0.9 |
| R2Z1 | R2Z4 | Florida St | Gonzaga | -5.9 | -5.5 | No choice | -0.4 |
| R2Z2 | R2Z3 | Texas A&M | Michigan | -2.9 | -2.5 | No choice | -0.4 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The model recommendations for round 3 are not as strong as in previous rounds. The recommendations in order of strength are:
Fringe picks:
# Append Final Scores of Games
round3 <- attachGameResults(round3, round3results)
# Determine ATS winner and Model's Accuracy ATS
round3 <- ATSresults(round3)
# Table of round 3 results and outcomes
resultsTable(round3)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| R2W1 | Villanova | R2W4 | West Virginia | 2.9 | 5.5 | West Virginia | 90 | 78 | Villanova | Incorrect |
| R2W2 | Purdue | R2W3 | Texas Tech | 0.6 | 1.5 | No choice | 65 | 78 | Texas Tech | NA |
| R2X1 | Kansas | R2X4 | Clemson | 8.2 | 4.5 | Kansas | 80 | 76 | Clemson | Incorrect |
| R2X2 | Duke | R2X3 | Syracuse | 4.3 | 11.0 | Syracuse | 69 | 65 | Syracuse | Correct |
| R2Y1 | Kansas St | R2Y4 | Kentucky | -2.3 | -6.0 | Kansas St | 61 | 58 | Kansas St | Correct |
| R2Y2 | Nevada | R2Y3 | Loyola-Chicago | -0.8 | 1.5 | Loyola-Chicago | 68 | 69 | Loyola-Chicago | Correct |
| R2Z1 | Florida St | R2Z4 | Gonzaga | -5.9 | -5.5 | No choice | 75 | 60 | Florida St | NA |
| R2Z2 | Texas A&M | R2Z3 | Michigan | -2.9 | -2.5 | No choice | 72 | 99 | Michigan | NA |
# Summary of all tourney results
entiretourney <- rbind(entiretourney, round3)
table(entiretourney$Accuracy)##
## Correct Incorrect Push
## 21 16 1The model went 3-2 ATS in the Sweet 16 and is now 21-16-1 throughout the tourney.
# Recording Round 3 Winners into Round 4
bracket <- advance_winners(round3)The model predictions for all round 4 games are on the rightmost column in the output table below.
# Append model predicted spreads to bracket
round4 <- predict_ncaa_round(4)
# Append actual spreads from sportsbook
round4 <- attachSpreads(round4, round4spreads)
# Determine who model would bet on ATS
round4 <- ATSchoice(round4)
# Output of model recommendations
predictionTable(round4)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| R3W1 | R3W2 | Villanova | Texas Tech | -5.1 | 7.0 | Texas Tech | -12.1 |
| R3Y1 | R3Y2 | Kansas St | Loyola-Chicago | -1.9 | 1.5 | Loyola-Chicago | -3.4 |
| R3Z1 | R3Z2 | Florida St | Michigan | -1.7 | -4.5 | Florida St | 2.8 |
| R3X1 | R3X2 | Kansas | Duke | -3.4 | -3.5 | No choice | 0.1 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The model recommendations for round 4 are Texas Tech, Loyola-Chicago, and Florida St.
# Append Final Scores of Games
round4 <- attachGameResults(round4, round4results)
# Determine ATS winner and Model's Accuracy ATS
round4 <- ATSresults(round4)
# Table of results
resultsTable(round4)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| R3W1 | Villanova | R3W2 | Texas Tech | -5.1 | 7.0 | Texas Tech | 71 | 59 | Villanova | Incorrect |
| R3X1 | Kansas | R3X2 | Duke | -3.4 | -3.5 | No choice | 85 | 81 | Kansas | NA |
| R3Y1 | Kansas St | R3Y2 | Loyola-Chicago | -1.9 | 1.5 | Loyola-Chicago | 62 | 78 | Loyola-Chicago | Correct |
| R3Z1 | Florida St | R3Z2 | Michigan | -1.7 | -4.5 | Florida St | 54 | 58 | Florida St | Correct |
# Summary of all tourney results
entiretourney <- rbind(entiretourney, round4)
table(entiretourney$Accuracy)##
## Correct Incorrect Push
## 23 17 1The model went 2-1 ATS in the Elite 8 and is now 23-17-1 through the entire tourney for a 57.5% success rate. Considering the sample size of 40 games, a 95% confidence interval for the true model’s accuracy is between 42.2% and 72.8%.
# Recording Round 4 Winners into Round 5
bracket <- advance_winners(round4)The model predictions for all round 5 games are on the rightmost column in the output table below.
# Append model predicted spreads to bracket
round5 <- predict_ncaa_round(5)
# Append actual spreads from sportsbook
round5 <- attachSpreads(round5, round5spreads)
# Determine who model would bet on ATS
round5 <- ATSchoice(round5)
# Output of model recommendations
predictionTable(round5)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| R4W1 | R4X1 | Villanova | Kansas | 1.9 | 5.0 | Kansas | -3.1 |
| R4Y1 | R4Z1 | Loyola-Chicago | Michigan | -6.8 | -5.5 | No choice | -1.3 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The only model recommendation for the Final Four is Kansas.
# Append Final Scores of Games
round5 <- attachGameResults(round5, round5results)
# Determine ATS winner and Model's Accuracy ATS
round5 <- ATSresults(round5)
# Table of results
resultsTable(round5)| StrongSeed | TeamName.x | WeakSeed | TeamName.y | Prediction | Spread | ModelChoice | Team.x.score | Team.y.score | ATSWinner | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|
| R4W1 | Villanova | R4X1 | Kansas | 1.9 | 5.0 | Kansas | 95 | 79 | Villanova | Incorrect |
| R4Y1 | Loyola-Chicago | R4Z1 | Michigan | -6.8 | -5.5 | No choice | 57 | 69 | Michigan | NA |
# Summary of all tourney results
entiretourney <- rbind(entiretourney, round5)
table(entiretourney$Accuracy)##
## Correct Incorrect Push
## 23 18 1Kansas was incorrect and the model is now 23-18-1 through the tournament.
# Recording Round 5 Winners into Round 6
bracket <- advance_winners(round5)# Append model predicted spreads to bracket
round6 <- predict_ncaa_round(6)# Append actual spreads from sportsbook
round6 <- attachSpreads(round6, round6spreads)# Determine who model would bet on ATS
round6 <- ATSchoice(round6)
# Output of model recommendations
predictionTable(round6)| StrongSeed | WeakSeed | TeamName.x | TeamName.y | Prediction | Spread | ModelChoice | Diff |
|---|---|---|---|---|---|---|---|
| R5WX | R5YZ | Villanova | Michigan | 2.9 | 7 | Michigan | -4.1 |
| Note: | |||||||
| A positive value in Prediction or Spread indicates TeamName.x being favored by that many points. A negative value indicates TeamName.y being favored by that many points. |
The model’s recommendation for the NCAA Championship is Michigan. It says Villanova is about 3 points better, rather than the 7 point spread.
Michigan did not cover the spread which is another loss for the model.
Over the course of the tournament, the model had a record of 23-19-1 ATS for 54.8% accuracy. That came in just under the 55% accuracy that was the objective of the model. I might revisit this to do further analysis and also will look for ways to improve the prediction model in next season’s tournament.