Design Predictive Model for RFID Tag Based Livestock Identification

and Monitoring System

Velmurugan Lingamuthu

, Tensae Endrias Zewdu

and Paul Mansingh

School of Computing Science and Engineering, VIT Bhopal University, Bhopal, India

Department of Computer Science, Ambo University, Ambo, Ethiopia

Department of Agricultural Extension & Economics, Vellore Institute of Technology, Vellore, India

Keywords: RFID, Dairy Yield, Cattle, Predictive.

Abstract: In the last recent years, the level of automation in the farming process has increased significantly. The key

component of these new techniques is live-stock identification and monitoring. As it is known, Ethiopia is

rich in its livestock sector but has never gained adequate profit from it. The basic problem is farm management

issues, attention given to it, and the livestock value chain inharmoniousness. The research aimed at automating

the traditional farm management practice using analytical processes. This paper uses an individual identifica-

tion of cattle intended for any farm using an ear tag embedded with radio frequency identification (RFID)

technology, where each cattle is tagged with an identifying number as a reference. The data was collected

from Alfa fooder & Dairy Farm P.L.C of years 2015 to 2020. The data then faded to a data mining software

to make a prediction based on the input data set. In this research work, an at-tempt has been made to apply

the comparative classification model predictive data mining techniques in the cattle livestock sector for the

milk, meat, and skin and hide quality yield products. Machine learning classification algorithms such as Naïve

Bayes, decision tree classifier and J48 classifier have been practiced. The overall model accuracy of Naïve

Bayes Net (94.24%) shows it has a better prediction.

1 INTRODUCTION

In Ethiopia, the government prioritized agricultural

production to promote economic growth overall, min-

imize poverty, and ensure food security. The agricul-

tural part of Ethiopia accounts for about 42% of the

GDP, more than 80% of the export, and 85% of the

employment opportunities (Ministry of Finance and

Economic Development, 2012). The largest popula-

tion of animals in Africa is currently in Ethiopia.

There are about 54 million cattle, 25.5 million sheep,

and 24,06 million goats estimated in the country

(Bekele et al., 2017). From the year 1995/96 to

2012/13, the cattle and shoat populations raised from

54.5 million to over 103.5 million, with an average

yearly growth of 3.4 million (Central Statistical Au-

thority, 2013). In 2026/27, the cattle, sheep, and goat

populations in the sedentary (people that do not travel

from place to place) areas of Ethiopia are estimated to

reach 75, 42.8, and 39.6 million heads, respectively.

The livestock sector majorly makes a significant

https://orcid.org/0000-0001-5535-8964

https://orcid.org/0000-0003-3423-8618

contribution to the economy as a whole. The sector

represents 19% of GDP and generates 16-19% of the

country’s foreign exchange earnings. It also accounts

for approximately 35% of agricultural GDP (or 45%

if indirect contributions are taken into account) (Cen-

tral Statistical Authority, 2013). Soon, domestic de-

mand for meat, milk, and skin and hide yield is ex-

pected to rise substantially with the speedily growing

population, increasing urbanization, and rising in-

comes.

Livestock identification in terms along the lines of

RFID and traceability are hot topics in today’s discus-

sions among livestock’s productivity. Let’s take one

sample type of RFID which comes with Livestock

show online software. First to identify an animal we

need to use a series of numbers that are unique which

are not used for another animal. This requires enough

digits to guarantee such uniqueness. now that we have

a unique number we can encode it to a metal chip

which is then embedded inside the plastic button that

is readable by a radiofrequency scanner called an

Lingamuthu, V., Zewdu, T. E. and Mansingh, P.

Design Predictive Model for RFID Tag Based Livestock Identiﬁcation and Monitoring System.

DOI: 10.5220/0012879300004519

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Emerging Innovations for Sustainable Agriculture (ICEISA 2024), pages 20-26

ISBN: 978-989-758-714-6

RFID reader these RFID readers typically comes in

the form of the hand wand or pocket device or

smartphone applications capable of reading the Elec-

tronic Identification (EID) number were brought near

to this button, this button can now be fixed to an ani-

mal in the form of ear tag to make things easier EID

ear tags are often combined with the more user-

friendly, human-readable and much shorter value

called visual ID or tag ID these short combination of

letters and numbers is easier to humans to read and

refer to than much longer and unique EID number.

2 LITERATURE REVIEW

Some newly upcoming mobile phones can also func-

tion as an RFID reader by their application software

installed where they can type in the information to ac-

quire. However, according to Cherinet Amsalu

(Cherinet Amsalu, 2015) just in case if the above ma-

terials could not be available in the Ethiopian market,

the barcode can also substitute since the working

mechanism of RFID and barcodes are related else

prototype can also be implemented as a simulation of

these physical devices. By developing an application

that provides RFID read event at a different location

and send the read event value to the application con-

tinuously while the Livestock Identification and

Traceability System (LITS) prototype is running on a

different machine. A list of programming, communi-

cation, databases and operating environments which

are appropriate for use in the implementation of a pro-

totype is provided below: two separate laptop or com-

puters, one for server and database storage and one

for client, SQL server 2010 and Microsoft Visual Stu-

dio 2008. The simulator function along the lines of a

chip or reader, even if the RFID chip device is not

located. The database (numeric, alphanumeric,

length, address and cattle full information) subse-

quently be built and recorded on a coded and format

basis. The data dictionary then be compiled.(Elec-

tronic identification for sheep and goats, 2024)

According to Cherinet (Cherinet Amsalu, 2015) to

fulfill a master’s thesis he designs and develops an

electronic cattle identification and traceability system

in Ethiopia, Borena Zone using a simulator as one PC,

which serves as a client and another PC as a server.

But this has not been implemented and tried to ad-

dresses the prediction of dairy quality products. De-

sign predictive model for RFID tag-based livestock

identification and monitoring system is a title chosen

after researching for similar papers and works done in

the different countries. The agriculture sector and ac-

ademic areas have indicated far more related issues

however, there was no such research done in our

country. The related works were done concerning in-

dicating the benefit of having a traceability and mon-

itoring system concerning RFID tags. No predictive

model using data mining techniques with RFID tags

has been designed.(Dogan, 2016)

For tracking and tracing of animals’ databases are

used. In these databases, individual animal identifica-

tion is linked to owner information and possibly other

information. The owner of the database can be a gov-

ernment or private oriented organization (or even it is

possible to maintain databases on owner e.g. farmer

level). A country may use several databases e.g. one

for companion animals, one for pigs, one for sheep,

one for goat, another for cattle, etc. Different organi-

zations can be responsible for different databases.

(ECDGHC, 2009)

There are four identification methods for cattle in

Botswana: ear markings, warm-iron branding, tradi-

tional ear markings (usually plastic) and bolus rumen,

which were introduced more than 11 years ago. These

methods of recognition are used concurrently. LITS

using rumen bolus had many problems, leading the

Botswana government resolve on 1 January 2013 to

replace it with electronic ear tags. It appears that most

of the challenges to LITS implementation are internal

processes that should have been addressed instead of

dumping the bolus system which offers some degree

of greater security than electronic ear tags. In differ-

ent researches, it has been shown that bolus has a high

retention rate and is tamperproof compared to elec-

tronic ear tags which can be easily removed, lost, or

tampered with.

Research Questions:

 What are the major determinants factors that

are used for constructing the predictive mod-

els?

 How machine learning models can be imple-

mented as predictive models for milk, meat,

and skin and hide quality yield products pre-

diction.

3 STATEMENT OF THE

PROBLEM

Ethiopia exports a large number of livestock products

to the middle east and other international markets.

However, the absence of an automated cattle manage-

ment system poses a great risk for further growth and

sustainability of the livestock trade. Some of these

challenges include not recording the cattle

Design Predictive Model for RFID Tag Based Livestock Identiﬁcation and Monitoring System

information appropriately and timely such as the age

of the cattle, the owner of the cattle, breed of the cat-

tle, sex of the cattle, body weight, blood level, health

status, and so on. So, need RFID like technique to

monitor and capture information from the cattle farm.

Also, analysis of livestock data that are captured

through RFID devices enables the farmers and agri-

cultural decision makers to utilize the benefit of the

emerging technologies and betterment in agricultural

field.

4 OBJECTIVE

The general objective of this research is to design a

predictive model for RFID tag-based livestock iden-

tification and monitoring systems that enable better

management of livestock milk, meat, and skin and

hide product.

 Specific Objectives

 To prepare a data set which are gathered

with the use of RFID technology to

make predictive system.

 Develop a predictive model using the

dataset of cattle and evaluate the pro-

posed design.

 Develop a suitable classification ma-

chine learning models for cattle meat,

milk, and skin and hide yield prediction.

5 MATERIALS AND

METHODOLOGY

In this paper, a predictive model for RFID tag-based

for cattle in Ethiopia was designed. Mainly this is

quantitative research in that data about livestock spe-

cifically cattle without RFID tags are collected. And

since this technology is new to our country and yet no

cattle had been tagged with RFID, we analyzed the

data by assuming the cattle’s data is collected with the

enhancement of RFID, then that data is analyzed us-

ing data mining techniques, particularly python to

come up with a predictive model to understand the

effectiveness of the RFID tags in increasing the cattle

milk, meat, and skin and hide quality yield products.

The main source of the data used to undertake this re-

search was cattle’s actual data taken from Alfa fooder

& Dairy Farm P.L.C which is collected with the use

of RFID technology. In this regard, the datasets were

found in softcopy which includes many more attrib-

utes to be taken, and some are hardcopy format with

7000 records and 17 variables. As we discuss with

different veterinarian experts, we identified the basic

problem domain and factors which can identify to

predict. The researchers first encoded all the data in

an Excel format and each record contains the most

relevant information about the cattle.

The processes like data cleaning, data prepro-

cessing and attribute selection was done using the

software python. The data set after the initial data col-

lection is as shown in the table 1.

Out of the 17 attributes of the original data set, at-

tributes (including the class attribute) which are be-

lieved by the domain experts to have significant con-

tribution in predicting milk, meat, and skin and hide

yield product of cattle’s, which is the focus of this re-

search, have been selected.

Table 1: Set of attributes

S.N

Attribute

Name

Data Type Descrip-

tion

1 ID Alphanu-

meric

The key that

identifies

the cattle

uni

uel

2 Year Number The year of

the data

capture

3 Place Text Exact resi-

dence of the

data col-

lecte

4 Owner

Name

Text Name of the

cattle owne

5 Sex Text Identifies

whether the

cattle are

male or fe-

male

6 Breed Text Identifies

originality

(ancestors)

of the cattle

7 Age in

Years

Number The exact

age of the

cattle

8 Body

Weight

Number The weight

of cattle in

kilograms

9 Blood Level Number The level of

blood pres-

sure

10 Feeding Text Grazing

type of the

cattle

whether in

the house

only or field

razin

ICEISA 2024 - International Conference on ‘Emerging Innovations for Sustainable Agriculture: Leveraging the potential of Digital

Innovations by the Farmers, Agri-tech Startups and Agribusiness Enterprises in Agricu

11 Health Text Health sta-

tus of the

cattle

12 Farm size Text The scope

of the far

13 Farm Sani-

tation

Text The clean-

ness of the

far

14 Use

Drug

Text The use of

medication

regularly for

the cattle

15 Farm Man-

agement

Text The moder-

nity of the

farm man-

agement

16 Farm

Housing

Text The general

farm hous-

standar

17 Number of

Calves

Number The number

of calves fe-

male cattle

has

The following are the parameters to extraction and se-

lection of features.

Records are evaluated and classified based on the

values of their attributes. Of course, some of the at-

tributes of a record may be irrelevant to the process

of classification and thus should be excluded. With

the help of area expertise the basic parameters we

choose that influence the prediction system is that ID,

year, place, owner, sex, breed, the cattle age in years,

body weight, blood level, feeding status, health status

of the cattle, farm size, farm sanitation, use drug/med-

ication, farm management, farm housing, and number

of calves the cattle has. The selection of attributes in-

cludes searching for the most successful sub-set of at-

tributes for prediction across all possible combina-

tions of attributes in the data.

The following are some parameters to extraction

and selection of features.

Milk:

• If sex is Male, print None.

• If sex is female, breed <> local, age in years

= 10, body weight >= 255, blood level >=

60, feeding is indoor, health status is excel-

lent, farm size is large, farm sanitation is yes,

use drug is yes, farm management is inten-

sive, farm housing is good, number of calves

is 4, print excellent milk.

• If sex is female, breed <> local, age in years

12, body weight >= 255, blood level >=60,

feeding is indoor, health status is very good,

farm size large, farm sanitation yes, use drug

yes farm management is intensive, farm

housing is good, number of calves is 5, print

very good milk.

• If sex is female, breed <> local, age in years

8, body weight >= 255, blood level >=60,

feeding is indoor, outdoor, health status is

good, farm size medium, farm sanitation

yes, use drug yes farm management is inten-

sive, farm housing is good, number of calves

is 3, print good milk.

• If sex is female, breed is local, age in years

6, body weight <= 255, blood level <=60,

feeding is indoor, health status is good, farm

size small, farm sanitation yes, use drug yes

farm management is semi intensive, farm

housing is good, number of calves is 2, print

satisfactory milk.

• If sex is female, breed is local, age in years

<=6, body weight >= 255, blood level >=60,

feeding is outdoor, health status is good,

farm size small, farm sanitation no, use drug

no, farm management is extensive, farm

housing is poor, number of calves is 2, print

poor milk.

Meat:

• If sex is Male, breed is Holiston, age in years

10, body weight >= 255, blood level >=60,

feeding is indoor, health status is excellent,

farm size large, farm sanitation yes, use drug

yes, farm management is intensive, farm

housing is good, number of calves is 0, print

Excellent Meat.

• If sex is Male, breed <> local, age in years

12, body weight >= 255, blood level >=60,

feeding is indoor, health status is very good,

farm size large, farm sanitation yes, use drug

yes, farm management is intensive, farm

housing is good, number of calves is 0, print

very good Meat.

• If sex is Male, breed <> local, age in years

8, body weight <= 255, blood level >=60,

feeding is indoor, health status is good, farm

size large, farm sanitation yes, use drug yes,

farm management is semi intensive, farm

housing is good, number of calves is 0, print

good Meat.

• If sex is Male, female, breed is all type, age

in years = 8, body weight <= 255, blood

level <=60, feeding is indoor, outdoor,

health status is satisfactory, farm size small,

farm sanitation yes, use drug yes, farm man-

agement is semi intensive, farm housing is

good, number of calves is <=4, print satis-

factory Meat.

Design Predictive Model for RFID Tag Based Livestock Identiﬁcation and Monitoring System

• If sex is female, breed is local, age in years

<=8, body weight >= 255, blood level >=60,

feeding is outdoor, health status is satisfac-

tory, farm size small, farm sanitation no, use

drug no, farm management is extensive,

farm housing is poor, number of calves is 4,

print poor Meat.

Skin and hide:

• If sex is Male, Female, breed is all type, age

in years <=10, body weight >= 255, blood

level >=60, feeding is indoor, health status is

excellent, farm size large, farm sanitation

yes, use drug yes, farm management is in-

tensive, farm housing is good, number of

calves is 0, print Excellent Skin and hide.

• If sex is Male, Female, breed is all type, age

in years <=10, body weight >= 255, blood

level >=60, feeding is indoor, health status is

Very good, farm size large, small, farm san-

itation yes, use drug yes, farm management

is intensive, farm housing is good, number

of calves is 0, print Very Good Skin and

hide.

• If sex is Male, Female, breed is all type, age

in years =10, body weight >= 255, blood

level <=60, feeding is indoor, outdoor,

health status is Good, farm size large, small,

farm sanitation yes, use drug yes, farm man-

agement is intensive, farm housing is good,

number of calves is <=4, print Good Skin

and hide.

• If sex is Male, Female, breed is all type, age

in years <= 10, body weight <= 255, blood

level <=60, feeding is indoor, outdoor,

health status is satisfactory, farm size large,

small, farm sanitation yes, no, use drug yes,

farm management is semi intensive, farm

housing is good, number of calves is <=4,

print satisfactory Skin and hide.

• If sex is Male, Female, breed is all type, age

in years <8, body weight <= 255, blood level

<=60, feeding is indoor, outdoor, health sta-

tus is satisfactory, farm size small, farm san-

itation no, use drug no, farm management is

extensive, farm housing is poor, number of

calves is <=4, print poor Skin and hide.

A comparative model algorithm is a powerful class of

machine learning algorithms that compare the predic-

tions from multiple models. The benefit of using py-

thon for applied learning machine is that it makes

available so many different comparative machine

learning algorithms. A popular advantage of a com-

parative model is that they allow one to compare the

prediction dairy qualities of multiple models. Thus,

with different models, we can calculate the milk,

meat, and skin and hide products.

The 10-fold cross-validation is the data is divided

randomly into 10 parts in which the classes are repre-

sented in approximately the same proportions as in

the full dataset (stratification). In turn, each part is re-

tained and the algorithm is trained on the other nine

parts, and the error rate on the holdout set is deter-

mined. Finally, the sum of the 10-error yield is an ap-

proximation of the total error. The quality of the data

measured in data mining is the error rate. This error

rate measured in classification accuracy, the standard

accuracy measurement in data mining is precision and

recall.

The machine learning models like Naive Bayesian

classifiers, J48 Algorithm, and Decision Tree & sta-

tistical model Bayesian classifiers were used as a pre-

dictive model.

The following figure 1 depicts the workflow of this

research work.

Figure 1: Shows the steps of a comparative classification

model

The models were implemented and it shows that

Naïve Bayes Net classifier outperformed other mod-

els with the highest accuracy of 94.24%. The follow-

ing table shows the performances of different classi-

fier models.

ICEISA 2024 - International Conference on ‘Emerging Innovations for Sustainable Agriculture: Leveraging the potential of Digital

Innovations by the Farmers, Agri-tech Startups and Agribusiness Enterprises in Agricu

Table 2: Summary of Performances by different Classifica-

tion Models

Perfor-

mance

Testing

Naïve

Bayes Up-

datable

Naïve

Bayes

Net

Deci-

sion

Tree

(HOE)

J48 Deci-

sion

Stump

Accu-

racy %

85.95 94.24 92.87 94 21.61

Av. Pre-

cision

0.9 0.94 0.92 0.94 0.21

Av. Re-

call

0.9 0.94 0.92 0.94 0.21

Av.

True

Positive

0.9 0.94 0.91 0.94 0.21

Av.

False

Positive

0.025 0.14 0.02 0.015 0.19

Sensi-

tivity

0.9 0.94 0.92 0.94 0.21

Speci-

ficity

0.9 0.94 0.92 0.94 0.21

6 CONCLUSION

In this research, an attempt has been made to apply

the comparative classification model predictive data

mining techniques in the cattle livestock sector and

the milk, meat, and skin and hide quality yield prod-

ucts. To achieve this goal, the KDD standard data

mining methodology has been adopted and the python

data mining tool has been used to implement the clas-

sification algorithm such as Naïve Bayes, decision

tree classifier and J48 classifier has been practiced.

The data for this research is the cattle data of the year

from mid of 2015 up to mid of 2020 collected from

Alfa fooder & Dairy Farm P.L.C university research

center bureau. After pre-processing out of 7,000 rec-

ords, 7000 cattle records are remaining for the partic-

ular reason for this circumstance we have used the

replication method and used it for building the mod-

els. But 20 attributes were minimized to 17 attributes

after pre-processing. Various experiments are made

iteratively by making adjustments to the parameters

and using a different number of attributes to come up

with a meaningful output. The comparison of the

models using python’s experimenter showed that

there is a relatively better model prediction in the case

of Naïve Bayes Net of Naïve Bayes correctly identi-

fying the dataset. The overall model accuracy of Na-

ïve Bayes Net (94.24%) shows it has a better predic-

tion. The relatively better performance of the Naïve

Bayes Net algorithm can be attributed to the nature of

the data such as the handled missing values; the data

consistency etc. we also gained the information that

the majority class of excellent meat quality was 3082,

the class of satisfactory meat quantity was 2196, the

class of poor meat quality was 537 and the minority

class of very good meat quality was 531. From our

total dataset of 6 Year (7000datasets), the majority of

cattle livestock was mid of the 2015 Year up to mid

of 2020 years, 1181, 1167, 1142, 1195, 1158, and

1157 respectively. As we see from this to sum up eve-

rything that has been stated so far, more numbers

were used in 2018 and the last one is 2017.

Thus, the results obtained in this research have

proved the applicability of data mining in cattle live-

stock identification and monitoring system. More

specifically it provides valuable help in developing

new methods to increase dairy products, particularly

in the milk, meat, and skin and hide products.

REFERENCES

Ministry of Finance and Economic Development, “Ethio-

pia’s Progress towards Eradicating Poverty: An Interim

Report on Poverty Analysis Study (2010/11), Addis

Ababa, Ethiopia." World development 59 (2012): 461–

474. tr. Food Sci, 2012 -4(3), 373–380.

Bekele, G., Lamaro, M., Berhe, G., & Berhe, A., “Produc-

tion potential and preservation methods of hide and skin

in three selected districts of gambella region, south west

Ethiopia”, International journal of research Gran-

thaalayah. 2017- doi:10.5281/zenodo.345631.

Central Statistical Authority, “Agricultural sample survey,

2012/13 (2005 E.C.)”, Volume II: Report on Livestock

and livestock characteristics (Private peasant holdings)

(Statistical Bulletin 570). Addis Ababa. 2013.

Cherinet Amsalu, “Design and Development of Livestock

Identification and Traceability System in Ethiopia”, A

research project submitted to the graduate Programme

research office of HILCOE in partial fulfillment of the

requirements for the degree of Master of Science in

software engineering, May 2015, pp.50-94.

European Commission Directorate General for Health and

Consumers, “Study on the introduction of electronic

identification (EID) as official method to identify bo-

vine animals within the European Union” (2009).

Souza-Monteiro D M and Caswell J, “A 2004 - the econom-

ics of implementing traceability in beef supply chains:

Trends in major producing and trading countries”. Am-

herst: Department of Resource Economics, University

of Massachusetts, retrieved from http://peo-

ple.umass.edu/resec/workingpapers/docu-

ments/resecworkingpaper2004-6.pdf, last accessed on

February 16, 2019.

Smith G C, Tatum J D, Belk K E, Scanga J A, Grandin, T

and Sofos J N,”Traceability from a US perspective”,

Meat Science, 2005 71: 174–193.

Daudi E. Ekuam, “Livestock Identification Traceability and

Tracking -It’s Role in Enhancing Human Security”,

Disease Control and Livestock Marketing in IGAD Re-

gion" 2009, pp. 41-4

Design Predictive Model for RFID Tag Based Livestock Identiﬁcation and Monitoring System

Habib Dogan, “Use of Radio Frequency Identification Sys-

tems on Animal Monitoring”,

SDU International Journal of Technological Sciences,

vol.8, No.2, 2016.

Moreki J. C., Ndubo N. S., Ditshupo T. and Ntesang J. B.

“Cattle Identification and Traceability in Botswana”, J

Anim Sci Adv 2012, 2(12): 925-933.

Sergio Pavon, “Animal identification in the European Un-

ion”, Key principles of creation of national systems of

Identification and Traceability of farm livestock, ICAR

seminar, European Commission DG Health and Con-

sumer Protection (DG SANCO) , Animal health,2015.

Electronic identification for sheep and goats, Department of

Primary Industries and Regional Development, Gov-

ernment of Western Australia, January 2024.

ICEISA 2024 - International Conference on ‘Emerging Innovations for Sustainable Agriculture: Leveraging the potential of Digital

Innovations by the Farmers, Agri-tech Startups and Agribusiness Enterprises in Agricu