BANGALORE UNIVERSITY

JNANA BHARATHI CAMPUS, BANGALORE-560056

DEPARTMENT OF COMPUTER SCIENCE AND APPLICATIONS

A project report entitled

“MEDICAL INSURANCE COST

PREDICTION”

Submitted By
KIRAN

P03NK21S0705

Under the Guidance of

Dr. M.T.SOMASHEKARA
DEPARTMENT OF COMPUTER SCIENCE AND APPLICATIONS
TOWARDS
MAIN PROJECT
PRESCRIBED BY THE BANGALORE UNIVERSITY FOR THE IV SEMESTER
MASTER OF COMPUTER SCIENCE
FOR THE ACADEMIC YEAR

2022-2023
 BANGALORE UNIVERSITY
JNANA BHARATHI CAMPUS, BANGALORE-560056

DEPARTMENT OF COMPUTER SCIENCE AND APPLICATIONS

CERTIFICATE
This is to certify that KIRAN Register No: P03NK21S0705 have satisfactorily
completed the project work entitled “MEDICAL INSURANCE COST PREDICTION”
for partial fulfillment of requirement for the award of the degree Master of Computer
Application (MCA) awarded by Bangalore University for the year 2022-23.

Dr.M.T.Somashekara Dr.Muralidhara.B.L GUIDE

HOD

Examiners:
1)

2)
 DECLARATION

KIRAN, student of IV semester MCA of Department of Computer Science and

Application, Bangalore University, hereby declare that the dissertation entitled
“MEDICAL INSURANCE COST PREDICTION” has been independently
carried out by me at Department of Computer Applications, Bangalore University
and submitted during the academic year 2022-2023. Further the matter embodied in
the report is an original and Bonafede work done by me.
To our knowledge this dissertation has not been submitted to any other college or
university or published at any prior to this.

Place: Bangalore KIRAN

Date:
 ACKNOWLEDGEMENT

The successful presentation of this project is an acknowledgement of the immense

support extended by Bangalore University which has provided an opportunity to
fulfill the most cherished desire to reach my goal.

I would like to express my heartfelt thanks to our HOD Dr. MURALIDHARA B.L,
for extending all the facilities to carry out this project.

I would like to express out special thanks to my project guide Dr. M.T.
Somashekara, who has spent his precious time by guiding & encouraging me
throughout the development of the project.

Last but not the least; I am grateful to my parents, my friends & all the people who
have helped me directly or indirectly to make this project a success.

KIRAN
 ABSTRACT

Medical costs are one of the most common reoccurring expenses in a person's life. It
is general known that a person's lifestyle and numerous physical factors determine
the diseases or disorders they may get, and that these conditions determine medical
expenses. According to several research, there are several significant reasons that lead
to greater expenditures. smoking, age, and BMI are all factors in personal medical
care. The goal of this study is to examine and identify a link between personal medical
costs and other characteristics. Then, by generating linear regression models and
comparing them using ANOVA, we use the significant traits as predictors to forecast
medical expenditures. In our research, we discovered that smoking, age, and a higher
BMI all have a significant connection with higher medical expenditures, showing that
they are key contributors to the charges, and that the regression can predict the
charges with more than 75% accuracy. According to the World Health Organization,
personal medical and healthcare spending is growing faster than the global economy
This rise in spending has been related to a variety of factors, the most prominent of
which are smoking, ageing, and higher BMI. Using insurance data from diverse
persons with variables such as smoking, age, number of children, area, and BMI, we
hope to uncover a link between medical expenditures and other parameters
 TABLE OF CONTENTS

CHAPTER NO TITLE PAGE NO

1 INTRODUCTION 1-4
1.1 Overview
1.2 Motivation for project
1.3 Problem Definition
and Scenario
1.4 Organization of the
report
1.5 Summary
2 LITERATURE 5-8
REVIEW
2.1 Introduction
2.2 Paper 1
2.3 Summary
3 PROJECT 9-11
DESCRIPTION
3.1 Objective of the
project work
3.2 Existing system
3.3 Shortcoming of
Existing System
3.4 Proposed System
3.5 Benefits of Proposed
System
4 SYSTEM DESIGN 12-14

Department of Computer Science and Application

 4.1 Architecture Diagram
4.2 Sequence Diagram
4.3 Use Case Diagram
5 PROJECT 15-16
REQUIREMENTS
5.1 Hardware and
Software Specification
5.2 Technologies Used
6 MODULE 17-20
DESCRIPTION
6.1 Modules
6.2 Module 1
6.3 Module 2
6.4 Module 3
7 IMPLEMENTATION 21-28
7.1 User Interface
7.2 Implementation Steps
7.3 Implementation
Procedure
8 RESULT ANALYSIS 29-33
8.1 Result Obtained
9 CONCLUSION AND 34
FUTURE WORK
9.1 Conclusion
9.2 Future Work

Department of Computer Science and Application

 LIST OF FIGURES

FIGURE NO TITLE

4.1 Architecture Diagram

4.2 Sequence Diagram

4.3 Use Case Diagram

4.4 Activity Diagram

LIST OF ABBREVIATIONS

Abbreviation Expansion

LR Linear regression

ML Machine learning

BMI Body mass index

DFD Data flow diagram

Department of Computer Science and Application

 CHAPTER 1
INTRODUCTION

1.1 OVERVIEW
The expense of health care is rising every day. There is a need to
forecast health costs as the number of novel viruses infecting humans
grows. This form of forecasting aids governments in making health-
related decisions. People are also aware of the significance of health-
care spending. Machine Learning is a field that touches all aspect of
life. Machine learning models are also used in the health-care system
for a variety of health-related applications. We conducted a predicate
analysis on medical health insurance expenses in this study. We create
a model to forecast a person's medical insurance costs depending on
gender. The dataset comes from Kaggle and comprises 1338 rows of
data with the following attributes: age, gender, smoker, BMI, children,
region, and insurance charges. Medical information and expenditures
billed by health insurance companies are included in the data. To
forecast medical expenses, we used a variety of regression techniques
on this dataset. The Python programming language was utilized to
implement the project.

In 2015, the absolute consumption of medical services as a proportion

of GDP was 3.89 percent, according to the World Bank. Legal medical
usage amounts for barely
1% of GDP, down from 3.89 percent in 2015, and cash-based medical
use makes for 65.06 percent of current medical use. Over the previous
few decades, advances in clinical innovation have made it feasible to

1
Department of Computer Science and Application
 cure ailments. It deals with a subject that was formerly supposed to be
fatal. In any event, the expense of her therapy is so high that it is

unaffordable for someone in the white-collar class. A 5,000 rupee

floater plan for yourself, your spouse, and your children is anticipated
to cost 10,000 to 17,000 rupees per year, while a 5,000 rupee sickness
plan will cost 4,000 rupees for multiple people.. In any case, the
expense of their therapy is so high that it is practically impossible for
a white-collar worker to pay it. A 5,000 rupee floater plan for a family
will cost between 10,000 and 17,000 rupees per year, whereas a 5,000
rupee health insurance plan for multiple years will cost 4,000 rupees.
It ranges in price from a few rupees to 7,000 rupees. It'll take a year.
According to the calculations.

It then generates linear regression models and compares them using

ANOVA to employ an important function as a predictor for forecasting
medical expenditures. Our research discovered that smoking, being
older, and having a higher BMI were all linked to higher medical
expenditures. This demonstrates that these are the primary sources of
cost, and that regression can accurately estimate expenses with a 75%
accuracy rate.

1.2 Motivation for the Project

To project will predict the outmost medical expenses that will help the
candidate or patient to claim the medical insurance and can know the
estimated medication cost. Finding the estimated medicine expenses
for a treatment is difficult, so that we want to them by giving
estimation for medication. That will know the approximately. In this
project our objective in this research is to forecast medical prices using
the data we have. We will examine the work of many writers in the
2
Department of Computer Science and Application
 field of price prediction in the first few chapters of this study, as well
as offer detailed information on some of the strategies utilized in the
health care pricing. Later, we’ll suggest a new system’s design based
on Medicare payments statistics.

1.3 Problem Definition and Scenarios

In the current world, the medical expenses plays the vital role in life.
sometimes we need to pay huge amount for treatments. We are going
to build a system, which predicts the medical insurance amount from
the company based on certain parameters and which also provides cost
of treatment for specific disease.
With help of machine learning we are going to build a system, which
takes inputs from the user and predicts the medical treatment cost of
the disease and insurance claim amount. for this we are using linear
regression.

Linear regression: Predict the response variable's result using some

explanator factors.

1.4 Organization of the report

The chapter 1is details about Overview of the motivation for the
project and problem statement purpose of the project, this tell us how
the project is described
The chapter 2 is about the literature review on the various paper that
are studied and understanding of the project in more detailed way
and analysis of the system that are described

3
Department of Computer Science and Application
 The chapter 3 is mainly discussed on the Project goals, current
systems, existing system flaws, prospective systems, and suggested
system benefits.
The chapter 4 is discussed on the project description in which the
architecture diagram, sequence diagram use case diagram, activity
diagram.
The chapter 5 is all about the requirements which is hardware and
software requirements specification and technologies used in the
project.
The chapter 6 is all about the module description used in which the
testing stage is classified into stages such as unit test, integrated
testing and coding of the parts that are implement in the system and
the result part of the system and also the bread bord testing which
helps for correct working of the model.
The chapter 7 is all about user interface and system that will explain
detailed about the project.
The chapter 8 is all about which we got results obtained and testing.

The chapter 9 is all about the conclusion of the project in which how
the project is concluded and what we have done in this project and
also about the future enhancement of the project what will be the
extension of this project for later development This the organization
of thesis what we have discussed in the following chapter this brief
understanding on each and every chapte

4
Department of Computer Science and Application
 CHAPTER 2
LITERATURE REVIEW

2.1 Introduction
Machine learning is a technique that allows computer to learn from
past data and anticipate fresh samples. Machine Learning models may
be used in any sector. Medical records are likewise not exempt from
machine learning. For numerous years, the medical industry has used
models in various settings. Many of the studies used machine learning
approaches to forecast medical costs B. Nithya [1] et.al In Predictive
Analytics in Health Care, machine learning models were used. For
predictive analysis, they used a variety of supervised and
unsupervised models. They also claimed that machine learning tools
and techniques are crucial in health-care sectors, and that they are
exclusively employed in the detection and prognosis of various
malignancies. Ahuja Tike[2] et.al applied hierarchical decision trees
for the medical price prediction systems. Their experiments showed
that the price prediction system achieves high accuracy. Moran et al.
[3] utilized linear regression techniques to anticipate Intensive Care
Unit (ICU) expenses and utilize understanding socioeconomics, DRG
(Diagnostic Related Group), length of stay in the clinic, and a couple
of others as highlights. Gregory [4] et.al applied various regression
models for analyzing medical costs in the health care system. They
mainly concentrated on reducing the bias in the cost estimates to
achieve good results. Dimitris Bertsimas[5] et.al applied different data
mining techniques which provided an accurate prediction of medical

5
Department of Computer Science and Application
 costs and represent a powerful tool for the prediction of healthcare
costs.

2.2 Medical Expense Prediction System using Machine

learning Techniques and Intelligent Fuzzy Approach (2020,
H. Chen Jonathan,
M. Asch Steven)

Prediction isn't a new concept in medicine. Clinical predictions based

on data are becoming commonplace in medicine., ranging Risk
categorization of patients in the critical care unit ranges from risk
scores to anticoagulant treatment (CHADS2) and cholesterol
medicine usage (ASCVD) (APACHE). You may easily develop
prediction models for hundreds of similar clinical questions using
clinical data sources and contemporary machine learning. These
approaches might be used for everything from sepsis early warning
systems to superhuman diagnostic imaging.
The real data source, on the other hand, has an issue. Unlike traditional
techniques, which rely on data from cohorts that have been thoroughly
prepared to prevent bias, new data sources are sometimes unstructured
due to the fact that they were developed for various purposes (clinical
care, billing, etc.). Patient self-selection, indication misunderstanding,
and inconsistent outcome data can all contribute to unintentional
biases and even racist programming in machine prediction. As a result
of this understanding, discussing the potential of data analysis to aid
medical decision-making isn't just wishful thinking.

6
Department of Computer Science and Application
 2.3 Predicting Days in hospital using health Insurance
Claims (2020, Yang Xie, C.W. Chang, Sandra Neubauer)

Identifying and managing patients most at risk within the health care
system is vital for governments, hospitals, and health insurers but they
use different metrics for identifying the patients they perceive to be at
most risk[1]. Hospitals focus on readmission rate [2], [3] and
cumulative risk of death during hospitalization [4]. Accurately
predicting these indicators could assist in allocating limited resources
and thus improve the hospital’s operational efficiency. Health insurers
are mostly concerned with insurance risk, because they agree to
reimburse health-related services in exchange for a fixed monthly
premium. Poor risk measure could result in exceeding a financial
budget. Therefore, one of the most obvious goals for health insurers is
to

Various predictive models have been developed to identify high-risk

customers by predicting health-care expenses [5]. Traditional
prediction models used demographic information and prior costs to
predict future costs [1]. More sophisticated models that incorporate
diagnoses [6], [7], drug claims [6], [7] and self-reported health status
data [8], have been shown to improve prediction performance. Zhao
et al. [6] reported a coefficient of determination (R2 ) of 0.168 when
both drug and diagnosis were used to estimate costs for the coming
year. Bertsimas et al. proposed two models: a decision tree model and
a clustering model [7], to improve the performance of an earlier model
based on classical regression models [6], [9]. Since hospitalization is
usually the largest component of health expenditure [10], a separate

7
Department of Computer Science and Application
 identification of subpopulations at higher hospitalization risk could
improve current underwriting processes and pricing methodology.

Moreover, insurance companies also manage insurance risk by using

specific interventions in different sub-populations (especially high-
risk groups) to minimize the resources they require [11]. Programs that
use case and disease management have been developed, which target
different sub-groups of customers, such as aged care programs,
chronic disease programs, and more recently telehealth programs, all
of which have been shown to improve health care outcomes. identify
high risk customers by predicting their health care expenditures.

2.4 Summary
This paper says about the various projects that are related to the Medical
price expense prediction using machine learning. Each one paper has
discovered new thing to the existing system and they implemented
various types of relations in them on this surveys and the other thing we
made new design to the existed system this is all about the literature
survey

8
Department of Computer Science and Application
 CHAPTER 3
PROJECT DESCRIPTION

3.1 Objective of the Project work

This device will help people to save time. As there will be no wastage
of time, the user will be satisfied. The WATERFALL MODEL, which
asserts that the stages are structured in a linear manner, Is essentially
being foll0wed First and foremost, Aa feasibility assessment has been
c0mpleted. After that the requirements analisis and project planning
may c0mmence. If an existing system requires changes or the
installation of a new module, an analysis of the existing system can be
utilized as a starting point. After the requirements analysis is
completes . the design phase begins . followed by the coding phase.
Tesing is a accomplished after the progrsmming is completed.
Requirement anlaisis, project planning, system interated and testing
are the activities conducted in this approach in a software
development project. The linear sequence of these action is crucial in
this case. The phase ends, and the output of one phase becomes the
input of the next.

9
Department of Computer Science and Application
 3.2 Existing System
• In sample sizes ranging from small to large, statistical approaches
(E.g., Lenear regresion) suffer due to the zsero point spike and skewed
distibutionof health care expenses with a strong right-hand tail.

• To address this issue, advanced methods have been proposed, such as

The data source, on the other hand, has a flaw. Unlike traditional
techniques, which rely on data from cohorts that have been

• thoroughly prepared to eliminate bias, new data sources are

sometimes unstructured since they were developed for various
purposes (clinical care, billing, etc.). A range of issues, ranging from
patient self-selection to indication uncertainty and inconsistent
outcome data, can induce unintentional biases and even racist
programming in computer prediction. As a result of this
understanding, speculating on the potential of data analysis to aid
medical decision-making isn't farfetched.

10
Department of Computer Science and Application
 3.3 Proposed System
• The Medical information and expenditures billed by health insurance
companies are included in the data. It has 1338 rows of information
with thw following columns: age, gender, BMI, illnesses, smokers,
and insurance costs.

• In these features, insurance charge is a dependent variable and the

remaining features are called independent variables.

• In regression analysis, we need to predict the value of the dependent

variable using independent variables. First, we collected the dataset
and applied various data preprocessing methods.

• Data preprocessing is a technique in which we can remove missing

values in the data. Because of these missing values, it is not possible
to apply machine learning algorithms. After removal of missing
values, we need to apply label encoding, one hot encoding data to the
categorical features.

11
Department of Computer Science and Application
 CHAPTER 4
SYSTEM DESIGN

4.1 Architecture Diagram

The diagram represents the implementation of the project that has done step by step
which has data preprocessing, cleaning, training and testing of the data as well. Our
projects procedure was as follows: first, we had to open the files with the Jupiter
notebook.

Price Prediction
System

Different techniques
proposed medical used by other
prediction systems people to predict
prices

Dataset infomation
Different Machine
and basic
Learning Techniques
architecture

Data pre-processing

Linear Regression
implementation

Results

Fig 4.1 Architecture Diagram

12
Department of Computer Science and Application
 4.2 Sequence Diagram
A sequence diagram depicts item interactions in chronological order.
It illustrates the scenario's objects and classes, as well as the sequence
of messages sent between them in order to carry out the scenario's
functionality. In the Logical View of the system under development,
sequence diagrams are often related with use case realizations.

Fig 4.2 Sequence Diagram

13
Department of Computer Science and Application
 4.3 Use case Diagram
Use case diagrams reflect the user's engagement with the system at the
most basic level by depicting the relationships between the user and
the many use cases in which the user participates. Use case diagrams
may be used to depict a variety of system users and use cases, and are
frequently accompanied by other diagrams. External entities that
engage with the system are referred to as actors. Circles or ellipses are
used to depict the use cases.

Open interface

Import libraries

Load datasets

User System
Pre processing

training & testing

Result

Fig 4.3 Use case Diagram

14
Department of Computer Science and Application
 CHAPTER 5
PROJECT REQUIREMENTS

5.1 Hardware and Software Specification

5.1.2 Hardware Requirements:

➢ System: Pentium IV 2.4 GHz

➢ Hard Disk:40GB
➢ Floppy Drive: 1.44Mb
➢ Monitor: 15VGA Color
➢ Mouse: Logitech
➢ Ram:512Mb

5.1.3 Software Requirements:

➢ Operating system: Windows XP/7

➢ Coding Language: python
➢ IDE: Anaconda Navigator

15
Department of Computer Science and Application
 5.2 Technologies Used
Software requirements specifications (SRS), often known as
software specifications, are a precise description of the behaviour of
the system in development. It provides a collection of scenarios that
define all of the software's interactions with the user. SRS also covers
non-functional requirements in addition to use cases. Nonfunctional
requirements are constraints on a system's design or execution. B.
System requirements specs & # 41; Performance requirements, quality
standards, or design limitations This is a set of data that contains all of
the system's needs. Business analysts, sometimes known as systems
analysts, are in charge of researching customer and stakeholder
business demands in order to uncover and provide solutions to
business problems. There are three components that must be included
in the project.
Business requirements define what must be supplied in order to
generate value.

16
Department of Computer Science and Application
 CHAPTER 6
MODULE DESCRIPTION

6.1 Modules
There are 5 modules in this project that have detailed explanation
below:
1) Dataset Information
2) Selection of features from the dataset
3) Data preprocessing

6.2 Module 1
Data set Information:
Our suggested system’s input dataset will be a dataset. That combines
two different datasets. A series of inpatient Medicare payment data
and a column of Zillow data will be displayed. be included in our final
input dataset. In the following part, we'll look at these columns in
further depth. The first dataset, Hospital-level payments to about
30,000 hospitals in the 100 most often billed Diagnosis Related
Groups are included in the Medicare payment data collection (DRGS).
The top one-hundred DRGS account for 60% of total inpatient
Medicare payments. expenditures and accounts for 7 million
discharges. Each row in the payment dataset comprises 10 columns,
as seen in the preceding section. Each of these columns denotes a
characteristic of machine learning. A feature is a significant attribute
that influences the prediction variable under consideration. Every
problem under investigation has a collection of independent
characteristics that aid in the construction of an accurate machine
learning
17
Department of Computer Science and Application
 6.3 Module 2

Feature Selection from the dataset

Selecting the key features from a large range of features that are more
relevant and building a strong model is a critical effort. As a result, we
will choose just those variables from our Medicare dataset that will
independently assist us in predicting medical prices. The location of
the provider is represented by columns such as provider address, ZIP
code, state, city, and hospital area referral description. As a result,
rather than examining all of them, we will just include one of them in
our feature set. We'll go with 'hospital region referral description'
because it's not as particular as a provider's location or city, but it's
also not as wide as a state. DRG Definition and Total Discharges are
two further independent characteristics we'll pick from medical
payment data. In addition to these features, the Medicare payment
dataset now includes a new feature: real estate values. These are the
prices of real estate in the hospital's immediate vicinity for the same
year as the medical data. This element, in our opinion, can potentially
be a prominent component in price prediction. The notion is that a
hospital's operating costs are factored into the price it charges. Among
the several expenditures that a hospital must face, one of the most
significant is the cost of real estate - the cost of owning or renting a
facility. Real estate expenses are also a proxy for other costs, in the
sense that if a location has high real estate costs, it is likely to have
higher costs elsewhere 19 Salary given to physicians, personnel, and
other categories are also included.

18
Department of Computer Science and Application
 6.4 Module 3

Data pre-processing:
The data that will be utilized to answer the problem is one of the most
significant aspects of machine learning difficulties. Data preparation
accounts for around sixty to seventy percent of the overall time spent
on a typical machine learning project. In order to get successful
outcomes, it is critical to have the proper data for the situation at hand.
In general, data preparation consists of selecting characteristics and
pre-processing those features. As a result, after selecting features from
a vast quantity of data, the following step is to pre-process those
features. Because the data is useless in its raw form. The goal of pre-
processing in this case is to make features appropriate for the machine
learning model we'll use. If the characteristics are set up correctly, the
model can produce better results. In addition, the data formats for
various models varies. There are machine learning models

The data pre-processing task includes following steps:

1. Data Integration
2. Data Cleaning
3. Data Transformation
We will describe each of these steps in details.

19
Department of Computer Science and Application
 Data Integration
Data integration is identifying the many data sources that will be
needed for processing and combining their information into one. This
stage is critical for any system that requires a large amount of data
processing to tackle the challenge at hand. In the sector, there are
several tools for integrating and combining data from various sources.
Such technologies can be utilized in situations when the amount of
data is large and there are several data sources.

Data Cleaning:
Many contaminants can be found in raw data. These contaminants can
have an impact on the ultimate result, especially when it comes to
machine learning difficulties. As a result, after the data has been
incorporated, it must be cleansed. Impurities such as incorrect entries,
irrelevant data, and inconsistent data are detected during data
cleaning. Eliminating these contaminants from records Data cleansing
may be accomplished using a variety of methods. Using automated
tools, personal interaction, and building scripts to programmatically
clean the data according to our needs are some of the data cleaning
strategies.

Transformation Data:
After cleaning data and integration data, the following is step to
convert the Clean data that has been incorporated into the system's
format. In most cases, data conversion entails transforming the target
data into the format needed by the source data.

20
Department of Computer Science and Application
 CHAPTER 7
IMPLEMENTATION

7.1 ALGORITHM USED: LINEAR REGRESSION

Linear regression is a kind of supervised machine learning. Carry do
the regression analysis. Regression models anticipate the target value
based on the independent variable. Its primary purpose is to anticipate
and identify relationships between variables. The sort of relationship
that is evaluated between the dependent and independent variables, as
well as the collection of independent variables that are used, differs
amongst regression models.
Linear regression is a statistical technique for predicting the value of
the dependent variable (y) based on the value of the independent
variable (y). A linear relationship between x (input) and y (output) is
identified as a result of this regression technique. The term "linear
regression" was coined as a result.
In the graph above, denotes personal salary and indicates job
experience. The passenger line is the finest option for our vehicle.
The following model is returned.
X: Schooling records (one variable enter parameter)
Y: Data that has been labelled (supervised learning)
While schooling the version, it suits the fine line to estimate the cost
of Y for a certain cost of. The version obtains the fine regression match
lane 1: by determining the fine 1 and a few numbers. Intersept 2: x's
C0-green

21
Department of Computer Science and Application
 We get the fit lane after we get the best one and two values. So when
we use our model to predict the value of y for the input value of x, it
will predict the value of y.

How can I change the values of 1 and 2 to achieve the greatest fit
line?
Cost Function (J):

By establishing an ideal regression trace, the model attempts to predict

the value in such a way that the error difference between the projected
and actual value is as little as feasible. As a result, it's critical to update
the values between 1 and to discover the best value that minimizes the
gap between the predicted y value (pred) and the actual y value (y).

A linear regression cost function y is the root mean square error

(RMSE) between the predicted y value (pred) and the actual y value
tre.

22
Department of Computer Science and Application
 Machine learning in regression in linear
Linear regression is one of the most widely used and straightforward
Machine Learning techniques. It's a statical predictive analytics
technique. Linear regression is used to predict sales, salary, age,
product price, and other continuous, real, or numeric variables.
As the name indicates, the linear regression process reveals a linear
relationship between a dependent and one or more independent
variables. Because linear regression displays a linear relationship, it
identifies how the dependent variable's value varies as the independent
variable's value changes. In a linear regression model, the
relationships between variables are represented by sloping straight
lines. Consider the diagram below.

Fig 7.1 A slop straight line representing the relationship

between the variables
Mathematically, we can represent a Linearr regression as:

23
Department of Computer Science and Application
 Y = Dependent variable in this case (target variable) X

stands for independent variable (predictive variable) a0 is

the intersection of two lines (additional degrees of

freedom) a1 = Coefficient of linear regression (scaling

factor for each input value).

= Error at Random

Training datasets for a linear regression model representation are the

values of the x and y variables.

Linear regression types

There are two different types of linear regression methods.

Simple linear regression is a kind of linear regression that uses a single
independent variable to predict the value of a numerically dependent
variable.

Numerous regression is a linear regression technique that predicts the

value of a numerically dependent variable using multiple independent
variables.

24
Department of Computer Science and Application
 Line of linear regression

Linear graphs that depict the connection between dependent and

independent variables are known as regression lines. There are two
sorts of associations that regression lines can represent:

Positive linear connection: If the dependent variable rises on the y-

axis while the independent variable increases on the x-axis, the
relationship is said to be positive

linear.

● Negative Linear Relationship:

If the dependent variable falls on the Y-axis while the independent
variable grows on
the X-axis, the connection is called a negative linear relationship.

25
Department of Computer Science and Application
 Assumptions of Linear Regression
Linear regression requires the following requirements. These are some
formal tests that you should run while developing your linear
regression model to guarantee that you receive the best results
possible from your dataset. A linear relationship between the
features and target:
In linear regression, the dependent and independent variables are
supposed to have a linear relationship.
Small or no multicollinearity between the features:
Multicollinearity refers to a significant connection between
independent variables. Due to multicollinearity, determining the true
relationship between the predictor and the target variables can be
challenging. As a consequence, the model denotes the absence of
minimum or multicollinearity in the feature or independent variable.

26
Department of Computer Science and Application
 Homoscedasticity Assumption:
When the error interval for all independent variable values is the same,
homoscadasty arises. There should be no discernible pattern
distribution of the data in a scatter plot using homoscefasciyu.
Normal distribution of error terms:
A normal distribution pattern is predicted for this linear regression
error. The confidence intervals will be too big or too narrow if the
error terms are not normally distributed, making estimating the
coefficient of determination difficult.
The qq chart may be used to verify this. The mistakes are evenly
distributed if the figure depicts a straight line with no deviations.
No autocorrelations:
The linear regression model, by error, does not imply autocorrelation.
The model's accuracy will be greatly diminished if the error terms are
correlated. If there is confidence between the residual errors,
autocorrelation might arise.
7.2 Implementation steps:
Step 1: Start
Step 2: Open interface
Step 3: Upload data sets
Step 4: Pre-processing
Step 5: Training and testing
Step 6: Enter constraints and submit
Step 7: result
Step 8: stop

27
Department of Computer Science and Application
 7.3 Implementation procedure:

The process of our project is as follows, Frist we had to open the files
by using the jupyter notebook. After opening we had to import the
libraries, now the user had to load the datasets. Now the system will
calculate the average charges for the each disease. The preprocessing
process will be started so now the data set will be preprocessed. The
dataset now will be trained first, next the testing will be performed.
Now the user need to enter the disease, smoker or non smoker, age.
etc. After submitting the constraints the result will be generated. The
result is the average cost and the insurance that can be claimed.

28
Department of Computer Science and Application
 CHAPTER 8
RESULT ANALYSIS

8.1 Results Obtained

The data set we're using is Kaggle's medical cost personal dataset,
which contains anonymous information about persons as well as
yearly insurance premiums. Age and BMI are both continuous
variables, whereas gender, smoking status, and geographic location
are all categorical. The dataset has no missing values, as can be seen
from the summary statistics below. The following are the
characteristics in this dataset Age Age of primary beneficiary person
at the time treatment Sex -Gender of insurance contractor
(female/male)
BMI -Body mass index (BMI) is a metric that compares the size of a
person's body to the health of that person's body.
Children -Number of children for insurance coverage.
Smoker -If your primary insurance company smokes.
Region -The beneficiary’s current residential address.
Charges -Health insurance companies bill for individual medical
expenses

29
Department of Computer Science and Application
 Fig 8.1 Dataset charges
This section deals with the result analysis of the project performance
of the model in the which we used the Linear regression algorithms.
In exploratory data analysis we have found the correlation between
the features and heat map to show correlation which we can see the
following observation. String correlation between charges and smoker
yes. Weak correlation between charges and age. Weak correlation
between charges and BMI. Weak correlation between BMI and region
_ southeast. Since the values for the weak correlations are less than
0.5 we can term them as insignificant and drop them. Which is shown
in below figure.

30
Department of Computer Science and Application
 Fig 8.2 correlation between features

We begin to predict the charges with the help of the other features. In
Model prediction our basic linear regression model predicting model
predicting the cost of treatment looks good. And the closely matching
results between training and test data means that our model is accurate.

31
Department of Computer Science and Application
 Fig 8.3 Heatmap

Fig 8.4 Training and testing accuracy

32
Department of Computer Science and Application
 Fig 8.5 Output

The above figure explains the graph and medical expenses that are
confirmed for the disease and also it also explains the insurance that
can be claimed by the patient. The patient can determine the cost that
is going to expense him for the disease from which he/she is suffering
and also it explains the patient that how much amount can be claimed
for the following disease based on the graph and expenditures.

33
Department of Computer Science and Application
 CHAPTER 9
CONCLUSION AND FUTURE WORK

9.1 Conclusion
We’ve looked at the fundamentals of the linear regression model, how
to use it o forecast charges, and how to compare anticipated and real
outcomes. I hope you found this post helpful and that you now have a
basic understaning of how a linear regression model works. For
estimating medical expenditures, we suggested a machine learning
approach.. We applied regression techniques Linear Regression and
observed that age, BMI are features that decide the dependent
variable. Out of all experiments, this model gave a better result.

9.2 Future Work

Work can be done in the future , making the system even more
scalable. We’re currently training and testing the algorithm with a
few thousand records. We csan attempt scaling the method for a
larger dataset with at least a million records in the future and see
how it performs . We can leverage distributed frameworks like spark
and Hadoop tomake the system scalable. These frameworks are
capable of effectively handling large amounts of data.

34
Department of Computer Science and Application
 REFERENCES

[1] A. Ravishankar Rao, Subrata Gardaí, Coumarate Dey, Hang Peng,

“Building predictive models of healthcare costs with open
healthcare data”,2020 IEEE
International Conference on Healthcare Informatics (ICHI) | 978-1-
7281-5382-
7/20/$31.00 ©2020 IEEE | DOI: 10.1109/ICHI48887.2020.9374348
[2] Pei Shen, “Factor Analysis of Medical Expenses of the Hepatitis A
patients in
Guangdong”, 2016 8th International Conference on Information
Technology in Medicine and Education.
[3] Ker-Tahj Shantung-Ming Yan and Pei-Wen Liu, “A Study on the
Annualized Medical Expense Prediction Model of the Bureau of
National Health Insurance --The
Application of the Grey Prediction Theory”, 2006 IEEE International
Conference on Systems, Man, and Cybernetics October 8-11, 2006,
Taipei, Taiwan.
[4] Sheng Yao Zhou, Run tong Zhang*, “A Novel Method for Mining
Abnormal
Expenses in Social Medical Insurance” Auckland University of
Technology. Downloaded on November 07,2020 at 17:01:50 UTC
from IEEE Xplore.
[5] Li Cheng, Sino Jalin Pan, “Semi-supervised Domain Adaptation
on Manifolds”,
IEEE TRANSACTIONS ON NEURAL NETWORKS AND
LEARNING SYSTEMS, VOL. 25, NO. 12, DECEMBER 2014.
[6] Yang Xia, Gunter Schreier, David C.W. Chang, Sandra Neubauer,
Ying Liu,
35
Department of Computer Science and Application
 Stephen J. Redmond, Nigel H. Lovell,” Predicting Days in Hospital
Using Health Insurance Claims”, IEEE Journal of Biomedical and
Health Informatics - DOI :10.1109/JBHI.2015.2402692.
[7] Shruti Kaushik, Abhinav Choudhury, Sayed Natarajan, Larry A.
Pickett, Varun
Dutti,” Medicine Expenditure Prediction via a Variance Based
Generative Adversarial Network”, 2018 IEEE International
Conference volume
[8] Anuja Tike, Sanket Tavarageri,”A Medical Price Prediction
System using
Hierarchical Decision Trees”,2017 IEEE International Conference on
Big Data
(BIGDATA

36
Department of Computer Science and Application
 APPENDEX A SAMPLE SCREEN SHOT

37
Department of Computer Science and Application
 38
Department of Computer Science and Application
 39
Department of Computer Science and Application
 APPENDIX B
SAMPLE CODE

import numpy as np import

pandas as pd import seaborn
as sns import
matplotlib.pyplot as plt
data_df=pd.read_csv("insuran
ce.csv")

# See sample data data_df.head() data_df.info()

data_df.isnull().sum() data_df.describe()
data_df.describe().columns df_num =
data_df[['age','bmi','charges']] df_cat =
data_df[['disease','sex','smoker','region']] df1 =
pd.get_dummies(df_cat) df1 data.corr()
sns.heatmap(data.corr(), cmap='RdBu')
data.corr()['charges'].sort_values() count, bin_edges
= np.histogram(data['charges'])
data['charges'].plot(kind='hist', xticks=bin_edges,
figsize=(20,12)) plt.title("Patient Charges")
plt.show()

from sklearn.linear_model import

LinearRegression from
sklearn.model_selection import train_test_split
from sklearn.metrics import
r2_score,mean_squared_error print('Please
40
Department of Computer Science and Application
 enter the disease: \n0-Ischemic heart disease,
or coronary artery disease\n1-Brain Stroke
\n2-Lower respiratory infections \n3-Chronic
obstructive pulmonary disease\n4-Trachea,
bronchus, and lung cancers \n 5-Diabetes
mellitus \n6-
Alzheimer’s disease and other dementias \n7-Dehydration due to
diarrheal diseases
\n8-Tuberculosis \n9-Cirrhosis\n') age, sex,
b, disease, smoker, region=0,0,0,0,0,0
age=int(input('Please enter your Age: '))
sex=input('Please enter your gender(m/f): ')
bmi=float(input('Please enter your Body
Mass Index: '))
print('Please enter the disease: \n0-Ischemic heart disease, or coronary
artery disease\n1-Brain Stroke \n2-Lower respiratory infections \n3-
Chronic obstructive pulmonary disease\n4-Trachea, bronchus, and
lung cancers \n5-Diabetes mellitus \n6-
Alzheimer’s disease and other dementias \n7-Dehydration due to
diarrheal diseases
\n8-Tuberculosis \n9-Cirrhosis') disease=int(input('Select the disease
appropriately: ')) smoker=int(input('Do you smoke? \nYes- 1\nNo -
0\nPlease enter your answer: ')) region=input('Please select the region
where you wanted to get medicated: \n0Hyderabad\n1-Banglore\n2-
Chennai\n3-Delhi\nPlease enter an apppropriate option: ')
x_new_predict={
"age":age,
"bmi":bmi,
41
Department of Computer Science and Application
 "disease_a":1 if disease==0 else 0,
"disease_b":1 if disease==1 else 0,
"disease_c":1 if disease==2 else 0,
"disease_d":1 if disease==3 else 0,
"disease_e":1 if disease==4 else 0,
"disease_f":1 if disease==5 else 0,
"disease_g":1 if disease==6 else 0,
"disease_h":1 if disease==7 else 0,
"disease_i":1 if disease==8 else 0,
"disease_j":1 if disease==9 else 0,
"sex_male":0 if sex=='f' else 1,
"sex_female":0 if sex=='m' else 1,
"smoker_no":1 if smoker!=1 else 0,
"smoker_yes":1 if smoker==1 else 0,
"region_Hyderabad":1 if region==0 else 0,
"region_Bangalore":1 if region==1 else 0,
"region_Chennai":1 if region==2 else 0,
"region_Delhi":1 if region==3 else 0,
}
y_new_predict=abs(lr.predict([list(x_ new_ predict.
values())])) print(' The average cost of medication is
',y_new_predict[0]*100) print('The estimated insurance that
can be claimed is ',y_new_predict[0]*70)

42
Department of Computer Science and Application