2018 3rd International Conference for Convergence in Technology (I2CT)

The Gateway Hotel, XION Complex, Wakad Road, Pune, India. Apr 06-08, 2018

Prediction Model: CO2 emission using machine

learning
Pooja Kadam Dr. Suhasini Vijayumar
Bharti vidyapeeth’s Institute of Management and Bharti vidyapeeth’s Institute of Management and
Information Technology, Information Technology,
Navi Mumbai, India Navi Mumbai, India
[email protected] [email protected]

Abstract- The paper provides insight of CO2 emission prediction algorithm to learn from experience. For example detecting the
model using machine learning. Traditionally researchers have spam mail automatically based on training. There are some
used statistical techniques such as regression, t-test, derivation, models developed using machines learning to predict CO2
ANOVA Test for prediction. The Machine learning provides emission.
different techniques to train the machine based on experience. In
this paper Supervised Machine learning regression technique is
used for the prediction of CO2 emission. In this paper an iterative The model developed by Chairul Saleh in 2015 using machine
and continuous improvement approach will be adapted to learning SVM algorithm to predict Carbon dioxide emission.
achieve successful results. The results are validated using Root The input parameter to the model is electrical energy and
Mean Square Error(RMSE) technique. burning coal which impact the quantity of CO2 emission. The
Index Terms— GHG, CO2, prediction, RMSE, carbon emission, data were collected from Alcohol Industry in order to training
machine learning and testing the model [2]. One more model was created by
Rui ZHAO in May 2017 which simplified the amount of CO2
I. INTRODUCTION emitted during production of raw milk. At different stages of
milk production including production of raw milk, dairy
One of the main reasons of climate change and changing earth processing, product transformation and packaging waste
temperature is emission of CO2. Presence of gas in atmosphere disposal. As per the results observed 1120g CO2 is emitted per
varies time to time. Some gases stay for few years and some liter. Collecting raw milk at farm contributes 75.27% to the
thousands of years. Some gases are more hazardous than CO2 emission, dairy processing 15.45%, product transportation
others. 3.39% and waste disposal 5.89[4].
According to the survey conducted by Inter governmental
Panel on Climate Change the temperature of land and ocean is The subsequent model we reviewed for detecting Software
increased by average 0.85°C. Hence the emission of gases in defects using machine learning techniques Artificial Neural
the last three decades was more. It has decreased number of Network and Decision Tree. As the software project grows its
cold days and nights and also increased number of hot days and difficult and expensive to test and evaluate the defects in the
nights [1]. project. The author “Onur Kutlubay” has developed two steps
machine learning model for predicting software defects. Here
Carbon dioxide(CO2) is more meticulous for Green House along with which module is defected author also predicted the
effect. There are different factors for CO2 emission such as magnitude of defect in the module [5].
living organisms, fermentation, burning of coal, natural gas,
and oil, trees, wood products, solid waste. Carbon footprint has There are so many industries which emit CO2, one of the
become a buzzword widely. We can see the online website for industry called sugar industry also produces CO2. The boiler
the home of carbon foot printing, carbon calculators, CO2 machine for processing sugar uses different material that
reduction, carbon offsetting and caring for the climate. produces CO2.The researcher “Chairul Saleh” has developed
the model for predicting CO2 emission using Artificial Neural
Network for sugar industry. Here 90% of the data is used for
II. LITERATURE SURVEY training purpose and 10% of data for testing purpose. The
accuracy of result is validated using RMSE [6].
CO2 emission in huge quantity is the major issue worldwide.
There are lots of researches going on to control the emission of III. PROPOSED REGRESSION MODEL FOR CO2 EMISSION
CO2. There is an ablaze need of model through which we can
predict the CO2 emission and take the proper action to control The regression model designed in this paper will investigate
the CO2 emission. Machine learning is the process of making CO2 emission as shown in figure 1. To create the model data is

978-1-5386-4273-3/18/$31.00 ©2018 IEEE 1

collected from the World Bank from 1960 to 2014. The CO2 is
measured in kt(Kiloton). After collecting data, the data is
normalized. The model will predict the CO2 emission based on
previous years data. To cross-validate the model the evaluation
technique is used by dividing dataset into training data(70%)
and testing data(30%).

Figure 2: CO2 emission Data

Figure 1: Design Regression Model Using linear regression algorithm data is trained for predicting
CO2 emission.
To design prediction model first I have collected the data from As shown in figure 3 scattered dots are actual data and line
internet then transferred the raw data into proper format. After obtained represents the prediction. The data selected for
Data Cleaning to validate the model, cross validation is prediction is only for India from 1960 to 2014. For each year
performed by dividing dataset into training data (70%) and the average value of CO2 emission is considered.
testing data (30%). The model proposed in the experiment uses
linear regression algorithm for predicting CO2 emission. The
subsequent function is used in model to predict the values.
f(X) =w0 + w1 X Eq.1
To evaluate the model RMSE is calculated to check the
accuracy of result.
IV. RESULT ANALYSIS

In this work we have used Anaconda Application software for

predicting CO2 emission. The main objective of research is to
monitor CO2 emission from different industries with lower
error.
In this experiment error measurement was performed on 70% Figure 3: Results of CO2 emission historical data and prediction
training dataset and 30% testing dataset. If RMSE is
Finally the output of regression model is analyzed and in this
significantly very small, then we are overfitting the problem
research we have received RMSE 0.2557. This model is
that is bad. If RMSE is significantly larger then there is a
developed based on the data received.
problem in model. In current study we have kept the actual
RMSE values without manipulation. CONCLUSIONS AND FUTURE WORK
As we can see from figure2, it represents the actual data where
x axis represent the year and y axis represent the CO2 emitted The regression model in the paper is used to predict CO2
in kt for india. emission based on historical data. For better prediction model
trial and error approach was applied to obtain lower error. The
Root Mean Square Error (RMSE) value is 0.2557. The main
objective of experiment is to achieve low RMSE value. We can
conclude that a lower RMSE value is propositional to more
accurate results. In further research the same data can be used
with different machine learning algorithm.
REFERENCES
[1] V. Sanjo Jose1, Veerasamy Sejian2,3*, Madiajagan
Bagath2, Athira P. Ratnakaran1, Angela M. Lees3, Yaqoub A.
S. Al-Hosni3, Megan Sullivan3, Raghavendra Bhatta2and John

2
 B. Gaughan3,“Modeling of Greenhouse Gas Emission from AN ARTIFICIAL NEURAL NETWORK: THE CASE OF THE
Livestock” SUGAR INDUSTRY, RESEARCH ARTICLE.
[2] Chairul Saleh1 , Nur Rachman Dzakiyullah2 , Jonathan Bayu [7] Muhammad Hassan Latif, Hammad Afzal, “Prediction of
Nugroho3,” Carbon dioxide emission prediction using support Movies popularity Using Machine Learning Techniques”,
vector machine”, Materials Science and Engineering g 114 IJCSNS International Journal of Computer Science and Network
(2016) 012148 Security, VOL.16 No.8, August 2016.
[3] Rui ZHAO1 ,Yao XU2 ,Xiangyu WEN1 ,Ning ZHANG3 ,Jiapei CAI1 [8] Mikiko Kainuma, Yuzuru Matsuoka, Tsuneyuki Morita, “The
“Carbon footprint assessment for a local branded pure milk AIM ⁄ ENDUSE Model and
product: a lifecycle based approach”, Food Sci. Technol, ahead Case Studies in Japan”, National Institute for Environmental
of print Epub May 08, 2017 Studies, Onogawa, Tsukuba, 305-0053, Japan
[9] Ishita Bhakta, Arkaprabha Sau, “Prediction of Depression
[4] FOSTER PROVOST, RON KOHAVI,“Guest Editors’ among Senior Citizens using Machine Learning Classifiers”,
Introduction: On Applied Research in Machine Learning”, International Journal of Computer Applications (0975 – 8887)
Machine Learning, 30, 127–132 (1998) °c 1998 Kluwer Volume 144 – No.7, June 2016
Academic Publishers, Boston. Manufactured in The [10] Ahmad LG*, Eshlaghy AT, Poorebrahimi A, Ebrahimi M and
Netherlands. Razavi AR, “Using Three Machine Learning Techniques for
[5] Onur Kutlubay, Mehmet Balman, Doğu Gül, Ayşe B. Bener, “A Predicting Breast Cancer Recurrence”, Ahmad et al., J Health
Machine Learning Based Model For Software Defect Med Inform 2013, 4:2
Prediction”, [11] Peter Harrington, “Machine Learning in Action”, by Manning
[6] Chairul Saleh1 , Raden Achmad Chairdino Leuveano2 , Mohd Publication Co.
Nizam Ab Rahman3 , Baba Md Deros4 , Nur Rachman [12] Tom M. Mitchell, “Machine Learning”, Tata McGrawHill
Dzakiyullah5, PREDICTION OF CO2 EMISSIONS USING