Introduction

• This app predicts the probability of India winning a cricket match using multinomial logistic regression trained on historical match results

• The model takes three user-selected inputs namely, opponent, venue, and format (Test / ODI / T20), and returns the predicted probability of a Win, Loss, or Draw

• The model is simplistic in nature, considering only a few key variables. Additionally, it is trained on historical data. Past performance is not a guarantee of future results

ML algorithm used and evaluation

The app uses a multinomial logistic regression model via the multinom() function from the nnet package. The model is trained on historical India match results, predicting outcome (Win/Loss/Draw) from opponent, venue, and format.

library(readxl)

## Warning: package 'readxl' was built under R version 4.5.2

library(nnet)

data <- read_excel("data/Ind vs opponent results.xlsx")
df <- subset(data, Result != "NR")
df$Result[df$Result %in% c("Tied","Draw")] <- "Draw"
df$Result <- factor(df$Result)
df$Opponent <- factor(df$Opponent)
df$Venue <- factor(df$Venue)
df$Format <- factor(df$Format)

model <- multinom(Result ~ Opponent + Venue + Format, data = df)

## # weights:  30 (18 variable)
## initial  value 1316.137522 
## iter  10 value 1026.379392
## iter  20 value 985.422744
## iter  30 value 980.993645
## final  value 980.975357 
## converged

summary(model)

## Call:
## multinom(formula = Result ~ Opponent + Venue + Format, data = df)
## 
## Coefficients:
##      (Intercept) OpponentEngland OpponentPakistan OpponentSouth Africa OpponentWest Indies  VenueHome VenueNeutral
## Loss    5.564669      -0.5569458        -1.551056            0.2923433          -0.9534112 -0.3348392     10.57441
## Win     4.810498      -0.2380381        -1.438593            0.4048800          -0.6323947  0.4819899     11.23608
##      FormatT20 FormatTest
## Loss  7.999503  -4.934941
## Win   8.695043  -5.078633
## 
## Std. Errors:
##      (Intercept) OpponentEngland OpponentPakistan OpponentSouth Africa OpponentWest Indies VenueHome VenueNeutral FormatT20
## Loss   0.7515430       0.2852758        0.3506165            0.4222167           0.2996347 0.2078663     205.1103  106.8301
## Win    0.7553691       0.2951833        0.3626136            0.4329006           0.3088540 0.2151487     205.1104  106.8301
##      FormatTest
## Loss  0.7219215
## Win   0.7234679
## 
## Residual Deviance: 1961.951 
## AIC: 1997.951

Results in graphical format

plot of chunk unnamed-chunk-1

Overall accuracy is moderate; the model performs best on Win predictions but struggles with Draw classification due to limited predictor variables.

Note on Data Collection

• Match results data was manually compiled from ESPN Cricinfo, covering India’s international matches across Test, ODI, and T20 formats against five major opponents: Australia, England, South Africa, Pakistan, and West Indies

• Each record contains four variables: Opponent, Venue, Format, and Result. Venue is categorized as Home, Away, or Neutral. No-result (NR) matches were excluded, and Tied/Drawn matches were merged into a single Draw category

• The dataset is limited in scope; only five opponents are included, and historical factors such as team composition, player form, and pitch conditions are not captured. Users should interpret predictions accordingly