2/12/25, 7:32 PM Homework 5 - Intro to Analytics Modeling
CS 7643 Homework 5 - Intro to Analytics
Modeling
2025-01-11
Question 8.1
for one of my earlier course at Georgia Tech, I built a predictive model to estimate the outcome of UFC fights using
Bayesian Statistics. I started with a linear regression approach and later transitioned to a nonlinear model for
better accuracy. The goal was to predict the winner between two fighters based on key attributes and statistics.
Below, I outline the steps and variables involved in this modeling process.
Linear Regression Approach
For the initial linear regression model, I focused on a combination of metrics to compare the fighters’ strengths in
areas like take-down prowess, striking ability, and age. The predictors included:
1. Take-down Prowess
This predictor compares each fighter’s ability to execute successful takedowns. Takedown success is a significant
factor in controlling the fight, especially for fighters with a wrestling background.
2. Striking Prowess
This measures a fighter’s ability to land significant strikes. Higher accuracy in striking generally correlates with
better offensive performance and greater control of the fight.
3. Age
The age difference between fighters is included because younger fighters tend to have more stamina and quicker
reflexes. Older fighters, on the other hand, may have more experience but can be physically less durable.
Refining the Model
To further improve the model’s predictions, I expanded it by including additional predictors:
4. Striking Accuracy
The percentage of strikes landed by a fighter. Fighters with higher accuracy can control the pace of the fight and
score points more effectively.
5. Take-down Defense
This predictor measures how well a fighter defends against take-down attempts. This is crucial when facing
wrestlers or fighters who heavily rely on ground control e.g. Khabib.
, 2/12/25, 7:32 PM Homework 5 - Intro to Analytics Modeling
6. Reach Advantage
The difference in arm reach between the two fighters. Reach advantage can affect a fighter’s ability to maintain
distance and land strikes from a longer distance, providing a significant edge in the fight.
7. Fight Experience (Total Fights)
The number of professional MMA bouts a fighter has participated in. More experience generally indicates better
adaptability to high-pressure situations and tougher opponents.
Model Description
By combining these variables in a linear regression model, I was able to estimate the likelihood of one fighter
winning over another based on their comparative advantages in various aspects of MMA. This model provided a
useful starting point for fight predictions. However, further refinements—specifically transitioning to a nonlinear
model—allowed for better accuracy. Some of the above mentioned predictors were not strictly linear, which is why
a nonlinear approach was more appropriate.
Question 8.2
1. Using LM for regression
Using linear modeling
Get the data:
data <- read.table("uscrime.txt", header = TRUE)
Separate into variables and response:
<- data[, 1:15] # The first 15 columns are the predictor variables
<- data[, 16] # The 11th column is the response variable (binary)
X_scaled <- scale(X)
data_scaled <- data.frame(X_scaled, Crime = Y)
Applying a linear model using lm:
model_linear1 <- lm(Crime~., data=data)
summary(model_linear1)
CS 7643 Homework 5 - Intro to Analytics
Modeling
2025-01-11
Question 8.1
for one of my earlier course at Georgia Tech, I built a predictive model to estimate the outcome of UFC fights using
Bayesian Statistics. I started with a linear regression approach and later transitioned to a nonlinear model for
better accuracy. The goal was to predict the winner between two fighters based on key attributes and statistics.
Below, I outline the steps and variables involved in this modeling process.
Linear Regression Approach
For the initial linear regression model, I focused on a combination of metrics to compare the fighters’ strengths in
areas like take-down prowess, striking ability, and age. The predictors included:
1. Take-down Prowess
This predictor compares each fighter’s ability to execute successful takedowns. Takedown success is a significant
factor in controlling the fight, especially for fighters with a wrestling background.
2. Striking Prowess
This measures a fighter’s ability to land significant strikes. Higher accuracy in striking generally correlates with
better offensive performance and greater control of the fight.
3. Age
The age difference between fighters is included because younger fighters tend to have more stamina and quicker
reflexes. Older fighters, on the other hand, may have more experience but can be physically less durable.
Refining the Model
To further improve the model’s predictions, I expanded it by including additional predictors:
4. Striking Accuracy
The percentage of strikes landed by a fighter. Fighters with higher accuracy can control the pace of the fight and
score points more effectively.
5. Take-down Defense
This predictor measures how well a fighter defends against take-down attempts. This is crucial when facing
wrestlers or fighters who heavily rely on ground control e.g. Khabib.
, 2/12/25, 7:32 PM Homework 5 - Intro to Analytics Modeling
6. Reach Advantage
The difference in arm reach between the two fighters. Reach advantage can affect a fighter’s ability to maintain
distance and land strikes from a longer distance, providing a significant edge in the fight.
7. Fight Experience (Total Fights)
The number of professional MMA bouts a fighter has participated in. More experience generally indicates better
adaptability to high-pressure situations and tougher opponents.
Model Description
By combining these variables in a linear regression model, I was able to estimate the likelihood of one fighter
winning over another based on their comparative advantages in various aspects of MMA. This model provided a
useful starting point for fight predictions. However, further refinements—specifically transitioning to a nonlinear
model—allowed for better accuracy. Some of the above mentioned predictors were not strictly linear, which is why
a nonlinear approach was more appropriate.
Question 8.2
1. Using LM for regression
Using linear modeling
Get the data:
data <- read.table("uscrime.txt", header = TRUE)
Separate into variables and response:
<- data[, 1:15] # The first 15 columns are the predictor variables
<- data[, 16] # The 11th column is the response variable (binary)
X_scaled <- scale(X)
data_scaled <- data.frame(X_scaled, Crime = Y)
Applying a linear model using lm:
model_linear1 <- lm(Crime~., data=data)
summary(model_linear1)
