CO4254 Machine Learning

Assignment - 01
Name: G.R.L Chandrasekera Index No. : 16/ENG/019
Topic: Compare the Linear Regression, Logistic Regression, and Poisson
Regression
Topic given in the assignment

Answer
 Supervised machine learning is widely used machine learning techniques that predict
future outcomes or event. Also supervised learning is classified into two categories as
regression and classification.
 Regression is a model that predicts continuous values. Also a logistic regression
algorithm is used for classification, while a linear regression algorithm is used for
regression. Also for modeling occasions where the results are count, Poisson regression
models work best.
 Linear Regression :
o Linear regression algorithm define a linear relationship between independent and
dependent variables. It predict the value of variable based on the value of another
variable.
o In this case, the variable that want to predict is called the dependent variable. The
variable that are using to predict the other variable’s value is called the independent
variable.
o There are two types of linear regression, those are simple and multiple linear
regression.
o When only one independent variable is present, simple linear regression should be
used to determine its linear relationship to the dependent variable. In contrast,
multiple independent variables are used in multiple linear regression to identify
relationship.
o When consider about equation of simple linear regression, where 𝑏0 is intercept, 𝑏1
Is coefficient or slop, x is the independent variable and y is the dependent variable.
y = 𝑏0 + 𝑏1 x

Page 1 of 4
 o When consider about equation of multiple linear regression, where 𝑏0 is intercept,
𝑏1 , 𝑏2 , 𝑏3 , 𝑏4 ….., 𝑏𝑛 are coefficient or slopes of the independent variables
𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ….., 𝑥𝑛 and y is dependent variable.

y = 𝑏0 + 𝑏1 𝑥1 + 𝑏2 𝑥2 ++ 𝑏3 𝑥3 ….. + 𝑏𝑛 𝑥𝑛

o When consider about graphical representation of linear regression, it is represented

by a straight line, also termed a regression line. This line displays the expected
score on "y" for each value of "x." Additionally, the distance between the data
points on the plot and the regression line reveals model flaws.

 Logistic Regression :
o Logistic regression is a statistical method used to predict the outcome of a
dependent variable based on previous observations. It is a commonly used
algorithm for solving binary classification problems.
o There are three types of logistic regression. Those are binary, multinomial and
ordinal logistic regression.
o Binary logistic regression means, when we have two possible outcomes, as an
example whether a person is likely to be infected with covid-19 or not.

Page 2 of 4
 o Multinomial logistic regression means, when have multiple outcomes, as an
example, predict whether someone may have the flu, an allergy, a cold or covid-
19.

o Ordinal logistic regression means, when the outcomes is ordered, as example, help
determines the severity of a covid-19 infection, sorting it into mild, moderate, and
severe cases.

o Logistic regression is applied across multiple areas and fields, additionally to

medical field, in financial industry, logistic regression can be used to predict if a
transaction is fraudulent or not. Also in marketing, ca be used to predict if a targeted
audience will respond or not.

o When consider about graphical representation of logistic regression, An S-shaped

curve is revealed by it. Here, the direction and steepness of the curve are affected
by changes in the regression coefficients.
y

 Poisson Regression :
o Poisson regression is intended for predicting numeric values, typically counts.
o Regression models called Poisson Regression use counts rather than fractional
numbers as the response variable. Count the births or victories in a football match
series, for instance. Additionally, the response variable values adhere to a Poisson
distribution.

Page 3 of 4
o Poisson regression would be useful in several scenarios;
- Modeling the number of colds associated with airplane flights
- Estimating the number of emergency service calls during an event
- Projecting the number of customer inquiries subsequent to a promotion
- Creating contingency tables

o Some of the issues with the normal model are resolved by the Poisson model. First,
the minimum value for the Poisson model is 0. Negative values will not be predicted
by it. As a result, it is perfect for distributions where the mean or most common
value is close to 0.
o Poisson regression has some limitations, Poisson regression may not perform well
in situations where the conditional variance is greater than the conditional mean.

o When comparing linear and Poisson regression,

Poisson regressions, in contrast to linear regressions, allow for non-continuous
dependent variables and the censoring of dependent variables at zero, but they also
make the firm assumption that the variance is equal to the mean.
o At some cases, Poisson regression is used instead of linear regression, because Use
of a Poisson regression model or one of its variations is an alternative. These models
differ from a standard linear regression model in a number of ways, such as the
skew, discrete distribution, and the limitation of predicted values to non-negative
numbers.
o When consider about applications of linear, logistic and Poisson regression, those
are used in different application sides.

o In specific, linear regression finds application in data science, business, finance, and
marketing. Also business insights, market analysis and financial risk assessment are
done using this regression.
o Logistic regression is used for medicine, credit scoring, customer behavior tracking,
hotel booking and text editing.

Page 4 of 4