Forest Fire Data Analysis Using Conventional Machine Learning
Algorithms
Cucu Ika Agustyaningrum
1
, Haryani
1
, Taufik Baidawi
1
, Wahyudin
1
, Siti Marlina
2
,
Artika Surniandari
1
and Sucitra Sahara
1
1
Fakultas Teknologi dan Informasi, Universitas Bina Sarana Informatika, Jakarta, Indonesia
2
Fakultas Teknologi Informasi, Universitas Nusa Mandiri, Jakarta, Indonesia
artika.ats@bsi.ac.id, sucitra.scr@bsi.ac.id
Keywords:
Algorithm, Conventional Machine Learning, Forest Fire, Method, Python.
Abstract:
A forest fire is a situation in which a forest is consumed by fire, damaging the forest’s products and causing
harm to the environment and the economy. Finding out how frequently forest fires occur is the aim of forest
fire prediction. The process of analyzing the data is therefore carried out using traditional machine learning
techniques utilizing the Random Forest, Decision Tree, Logistic Regression, Nave Bayes, and Multilayer Per-
ceptron methods. Knowing the accuracy and F1 score values allows for a comparison of this method using the
Python programming language. The test results showed that the multilayer peceptron approach outperformed
the Random Forest, Decision Tree, Logistic Regression, and Nave Bayes methods, with accuracy values of
86.70% and 87.93%, respectively, with a hidden layer size of 32.32. When compared to the other approaches
investigated, the value of the multilayer perceptron method is quite prominent. This research can help deter-
mine the probability of forest fires.
1 INTRODUCTION
Every nation on earth needs forests and forest ecosys-
tems to survive and develop socially, economically,
and environmentally. Forests are thought to be perma-
nently and seriously threatened by forest fires. Partic-
ularly for large fires, the detrimental effects of forest
fires persist for tens of years after they have burned.
The prevention of forest fires is one of the most cru-
cial issues, and it involves a variety of proactive mea-
sures in addition to retaliatory ones like fire suppres-
sion (Baranovskiy and Zharikova, 2014).
Forest fires, often known as wildfires, are one of
the major environmental issues because they have a
negative impact on the sustainability of forests, harm
the environment and economy, and hurt people. Mil-
lions of hectares (ha) of forests around the world suf-
fer damage each year as a result of the occurrence,
which is brought on by a variety of sources (including
human negligence and lightning strikes)(Cortez and
Morais, 2007).
In forest areas, which burn millions of hectares
annually and are responsible for loss of biodiversity,
soil quality, and CO2 capture. The vulnerability of
forests and their surrounding areas, that is, human set-
tlements and infrastructure, to fire is a major concern
for people in many of the world’s terrestrial ecosys-
tems. Increasing changes in socio-economic and cli-
matic processes leading to extensive modifications
to the natural environment and prolonged periods of
drought have placed strong demands on authorities
and decision-makers to delineate forest areas tempo-
rally and spatially in terms of vulnerability to fire.
Identifying areas with high or very high fire vulner-
ability is a must in order to successfully design a fire
management plan and allocate firefighting resources.
To this end, robust approaches and tools are needed to
enable managers and engineers to accurately predict
the timing, location, and extent of future fires. Im-
provements in techniques for predicting fire vulnera-
bility and describing forest areas according to differ-
ent levels of vulnerability can help forest managers
and policymakers achieve a better understanding of
fires, which facilitates the development of preventive
measures for fire-prone forests (Pham, 2020).
A forest fire is a situation in which a forest is con-
sumed by fire, causing harm to the forest products
that results in losses for the economy and the envi-
ronment. The greatest fires, which consumed over
2.6 million and 1.6 million hectares of Indonesian for-
32
Agustyaningrum, C., Haryani, H., Baidawi, T., Wahyudin, W., Marlina, S., Surniandari, A. and Sahara, S.
Forest Fire Data Analysis Using Conventional Machine Learning Algorithms.
DOI: 10.5220/0012441300003848
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 3rd International Conference on Advanced Information Scientific Development (ICAISD 2023), pages 32-37
ISBN: 978-989-758-678-1
Proceedings Copyright © 2024 by SCITEPRESS Science and Technology Publications, Lda.
est and land, respectively, occurred in Indonesia dur-
ing the ve-year period of 2015–2019. According to
data from the Ministry of Environment, Indonesia’s
protracted dry season and increasing sea levels con-
tributed to the two large fires that broke out around
that time of the year. About 29% of the two fires were
located in peatlands.
Numerous studies on the Indonesian forest fires
that have happened have been carried out, includ-
ing work by (Negoro et al., 2022) ”Fire Analysis in
Forests and Locations in Riau Province Using the
C4.5 Method” is the study’s title. Due to its signif-
icant advantages over competing algorithms, decision
tree classification method C4.5 is widely employed
(Manalu et al., 2021).According to the findings of
this study, forest and land fires in Riau Province are
heavily influenced by environmental conditions such
as humidity, weather, and wind speed. The results of
an analysis of low humidity (dry) and sunny weather
with high wind speeds can show a higher chance of
forest fires occurring. Meanwhile, cloudy weather
and high wind speeds can cause forest fires, although
the percentage is smaller. The C.45 algorithm method
is used in this study with the equation test, and the
results are compared to the confusion matrix.
Another study conducted by (Husen et al., 2022)
entitled ”Analysis of Forest Fire Prediction” Using
the Random Forest Classifier Algorithm, this research
develops the concept of a forest fire prediction sys-
tem, which will become one of the government’s pol-
icy references in issuing preventive policies. This re-
search conducts modeling using the Random Forest
Algorithm model on forest fire data from year to year
in Indonesian territory with the hope that it can assist
the government in preventing forest fires with its legal
policies and that existing analysis can be used by the
Weather Modification Technology Center (BBTMC),
which can help determine when weather modifica-
tions can be made.
Research conducted by (Yandi et al., 2021) with
the title ”Prediction of Forest Fire Hot Spots Us-
ing the SVM (Support Vector Machine) Regression
Model on Daops Manggala Agni Oki Forest Fire
Data, South Sumatra Province in 2019” The data
prediction method used is the SVM (Support Vec-
tor Machine) regression algorithm. machine) with
data (date, time, satellite, accuracy, district, subdis-
trict, humidity, and temperature). The study’s findings
yielded quite good analysis results, with an RSME
value of 2.1 and an R2 value of 0.83. where the most
hotspots result from the process in 2021. Meanwhile,
for 2022 data, the highest number of hotspots is in the
Cengal sub-district, with 571 hotspots. And to pro-
vide better data visualization, hotspot prediction re-
sults are visualized in the form of a heatmap.
Furthermore, research has been carried out by
(Ayuningtyas and Prasetyo, 2020) with the title Uti-
lization of Machine Learning Technology for Clas-
sification of Drought Risk Areas in the Special Re-
gion of Yogyakarta Using Landsat 8 Operational Land
Imager (OLI) Imagery. This research was conducted
to predict risk areas using satellite imagery with ma-
chine learning autocorrelation and artificial neural
network methods. The results of the Morans I anal-
ysis show that all vegetation indices and predictions
using ANN have positive autocorrection.
Research conducted by (Pratiwi et al., 2021) enti-
tled ”Classification of Forest and Land Fires Using the
Naive Bayes Algorithm” This study uses a dataset of
forest fires in Pelalawan Regency from 2015 to 2019
using the Naive Bayes method. Hot spots to be ana-
lyzed consist of temperature, humidity, rainfall, wind
speed, and class. The classification method using the
Naive Bayes algorithm can be used for prevention be-
fore forest and land fires occur.
From several related studies, in this study the au-
thors made comparisons with five algorithms, namely
Random Forest, Decision Tree, Logistic Regression,
Nave Bayes, and Multilayer Perceptron. The fore-
casting method that is often used in research is the
multilayer perceptron neural network (MLP) (Manalu
et al., 2021). The characteristics possessed by MLP
are its advantage in determining the weight value,
which is better than other methods; MLP can be used
without prior knowledge; the algorithm can be eas-
ily implemented; and it is able to solve both linear
and nonlinear problems (Manalu et al., 2021). From
these ve comparisons, the algorithm model that has
the highest level of comparison will be selected. From
the modeling results obtained, it is hoped that they can
assist the Indonesian government in taking the most
appropriate preventive actions against forest fires in
the future.
2 METHODS
A model for studying forest fires is developed on the
dataset, preprocessing, feature selection, and model
evaluation stages of the study technique. begin-
ning with gathering data via the UCI Machine Learn-
ing Repository website. After transforming the data
into starting data for preprocessing, features are se-
lected using Python, and feature selection is then vali-
dated using conventional machine learning algorithms
including Random Forest, Decision Tree, Logistic
Regression, Nave Bayes, and Multilayer Perceptron
methods, which are processed through training tests.
Forest Fire Data Analysis Using Conventional Machine Learning Algorithms
33
Five algorithms—Random Forest, Decision Tree,
Logistic Regression, Naive Bayes, and Multilayer
Perceptron—are being compared in the modeling
step. The data to be used is processed using a train
test with a train value of 0.7 and a test value of 0.2 be-
fore entering the modeling procedure. The next stage
is to use conventional machine learning algorithms to
examine the data generated by the Python program-
ming language in order to discover which techniques
achieved the best results for the dataset of forest fires.
After the data has undergone preprocessing, feature
selection, modeling, and testing, this is done.
This research will be carried out through several
stages as shown in Figure 1.
Figure 1: Research Stages.
2.1 Utilizing Research Techniques
The Forest Fires Data Set is used to apply the study
methodology in five stages, namely:
2.1.1 Datasets
In order to estimate the likelihood of forest fires, 517
data points with 13 attributes and 1 class were gath-
ered from the UCI Machine Learning Repository as
secondary data. It is possible to get around these is-
sues and get results that are simpler, faster, and more
precise by using a classification technique with the
maximum level of prediction and accuracy. By com-
paring the Random Forest, Decision Tree, Logistic
Regression, Nave Bayes, and Multilayer Perceptron
using Python as the study’s programming language,
it could produce results with a high degree of pre-
dictability and accuracy.
2.1.2 Preprocessing
517 pieces of data totaling 13 attributes and 1 class
were acquired for this study stage. These data will be
analyzed to create forecasts of forest fires based on
descriptions of the current attributes. The selection of
data, which entails assessing the qualities for which
the data type will be modified, is the first step in the
data preparation process. The data cleaning process
comes next after the data selection process has been
completed. In this procedure, you should try to look
for any missing values.
2.1.3 Feature Selection
It is applied to identify the features that have the great-
est impact on the data throughout the feature selec-
tion process. The modeling of conventional machine
learning algorithms then uses train and test on the data
distribution.
2.1.4 Modelling
The prediction process using the suggested method
carries out the modeling stage in a number of differ-
ent ways. To assess the degree of accuracy and f1
score in predicting forest fires, the proposed machine
learning algorithms use the Python programming
language and include Random Forest, Decision Tree,
Logistic Regression, Nave Bayes, and Multilayer
Perceptron approaches.
a. Machine Learning
Machine learning is the automatic recognition of sig-
nificant patterns in data. Computers can learn things
from people through machine learning. Without
any explicit programming, the computer can learn
to process the data that is given to it. Algorithms
for machine learning are used to train computers to
process data (Agustyaningrum et al., 2021).
b. Random Forest
The Random Forest concept is used to generate a large
number of correlated decision trees, with each deci-
sion tree acting as a set of models. Each decision tree
sets class predictions, and the final decision is based
on the maximum yield (Kabir et al., 2019). The ran-
dom forest classification method is based on a deci-
sion tree approach where attributes are randomly cho-
sen at each node to determine categorization. The
decision tree’s returned highest number of votes is
used to classify data (Ratnawati and Sulistyaningrum,
2019). Using a voting system (highest count) to com-
bine mutually independent classifiers (CARTs) from
the same distribution, random forest produces classifi-
ICAISD 2023 - International Conference on Advanced Information Scientific Development
34
cation predictions. Reduced correlation can lower the
outcome of random forest prediction errors, which is
a property of random forest (Sarofi et al., 2020). Ran-
dom Forest formula (Leonardo et al., 2020):
Entropy (Y ) =
i
P (Y )log
2
p(Y ) (1)
= Entropy (Y )
v
εvalues (a)
|
Y
v
|
|
Y
a
|
Entropy (Y
v
)
(2)
Information :
Y = case set
P(c—Y) is the ratio of grades in class Y to those in
class c.
Values(a) = Possible values when a is set.
Yv = subclass of Y with class v, which is related to
class a.
Ya = All values that correspond to a.
c. Support Vector Machine
A machine learning technique called support vector
machines operates under the tenets of structural
risk minimization (Saputra et al., 2022). Because it
requires specific learning objectives during training,
Support Vector Machine (SVM) is an integrated
(supervised) classification method (Nurachim, 2019).
The following is the support vector machine formula
(Zulfikar and Lukman, 2016):
similarity =
n
i=1
f (T
i,
S
i
)
W
i
(3)
Information:
T:A new case
S: cases in storage
n: the number of attributes
I: individual attribute between 1 and n
f: TRIBUTE similarity function between case T and
case S
W: weight assigned to the i-th attribute.
d. Logistics Regression
Regression with logistic data is part of the supervised
classification process. The use of this algorithm has
greatly expanded in recent years as has its popularity.
This curve is sigmoid. It belongs to the class of
logistic regression. To comprehend the mathematical
representation of the explanation, let’s start with a
straightforward linear regression formula (Shah et al.,
2020).
y = b0 + b1 x (4)
Thus, it has now been subjected to the sigmoid func-
tion, and the result is provided by the formula.
p =
1
1 + e
y
(5)
Now that one formula has been substituted for
another to get the value of y, we have our logistic
regression formula.
logit (S) = b0 + b1M1 + b2M2 + b3M3. . . bkMk . . .
(6)
where S denotes the likelihood of the presence of
interesting features. The predictor values are M1,
M2, M3, ... Mk. The intercepts of the model are bo,
b1, b2, b3, ... bk.
e. Multilayer Perceptron
A multilayer perceptron is a feed-forward artificial
neural network made up of many neurons connected
by their connecting weights. With an input layer,
one or more hidden layers, and an output layer, these
neurons are arranged in layers (Irfan et al., 2017).
f. Na
¨
ıve Bayes
Based on Bayes’ theorem for conditional probabili-
ties, Naive Bayes is an easy-to-understand algorithm.
This is done to categorize data based on how fre-
quently the data descriptor appears in the training
set. The Naive Bayes algorithm makes the supposi-
tion that all data are equally independent. The method
looks for dependencies between the training set’s fea-
ture set using this supposition (Agustyaningrum et al.,
2020).
2.1.5 Evaluation
The prediction process is carried out using traditional
machine learning algorithms in the assessment stage
in order to check the accuracy and f1 scores of success
and error rates. These methods include Random For-
est, Decision Tree, Logistic Regression, Naive Bayes,
and Multilayer Perceptron.
2.2 Method of Collecting Data
Primary data and secondary data are two categories
into which data collection techniques can be sepa-
rated. While secondary data is derived from schol-
ars who have already carried out related research, pri-
mary data is collected straight from the source. In Ta-
bles 1 and 2, a total of 517 records with 13 attributes
and 1 class attribute are drawn from the Forest Fires
Data Set, which was retrieved from the UCI Machine
Learning Repository for use in this study.
Table 1 Description of the attributes of the survival
dataset of forest fires.
Forest Fire Data Analysis Using Conventional Machine Learning Algorithms
35
Table 1: Description of the attributes of the survival dataset
of forest fires.
attribute
name
data type description
X Numeric x-axis spatial coordinate within the Montesinho park
map: 1 to 9
Y Numeric y-axis spatial coordinate within the Montesinho park
map: 2 to 9
month Category month of the year: ’jan’ to ’dec’
day Category day of the week: ’mon’ to ’sun’
FFMC Numeric FFMC index from the FWI system: 18.7 to 96.20
DMC Numeric DMC index from the FWI system: 1.1 to 291.3
DC Numeric DC index from the FWI system: 7.9 to 860.6
ISI Numeric ISI index from the FWI system: 0.0 to 56.10
temp Numeric temperature in Celsius degrees: 2.2 to 33.30
RH Numeric relative humidity in %: 15.0 to 100
wind Numeric wind speed in km/h: 0.40 to 9.40
rain Numeric outside rain in mm/m2 : 0.0 to 6.4
area Numeric the burned area of the forest (in ha): 0.00 to 1090.84
Table 2: Numerical Attributes and Categories of User Be-
havior Analysis.
Atribute
Name
Min. Value Max. Value STD
X 1 9 2.31
Y 2 9 1.23
FFMC 18.70 96.20 5.52
DMC 1.10 291.30 64.05
DC 7.90 860.6 248.07
ISI 0 56.1 4.56
temp 2.20 33.3 5.81
RH 15 100 16.32
wind 0.4 9.4 1.79
rain 0 6.4 0.3
area 0 1090.84 63.66
3 RESULTS AND DISCUSSION
With a total of 517 records and 13 attributes and one
class attribute, the secondary data required to make
forest fire predictions was collected from the UCI
Machine Learning Repository. According to Paulo
Cortez and Anibal Morais’ study on ”A Data Mining
Approach to Predict Forest Fires Using Meteorolog-
ical Data, which employed the SVM and Random
Forest methods, the attribute temperature of 9.95, RH
of 0.56, winds of 0.64, and rains of 2.45 also produce
a relativized temperature of 73.2%, 4.1% RH, 4.7%
wind, and 18% rain, which produced the best SVM
variance pattern.
The study’s findings consist of both qualita-
tive and quantitative information that was gathered
through calculations using the suggested model. All
of the data sets that were accessible were used for
this study. Research experiments and testing are con-
ducted by projecting data sets using traditional ma-
chine learning methods. This experiment will be run
on datasets that have been approved based on the out-
comes of preprocessing, feature selection, modeling,
and evaluation performed using the Python program-
ming language and the Google Collaboratory.
3.1 Preprocessing, Step Validation, and
Conventional Machine Learning
Algorithm Models
The following values were generated by study em-
ploying preprocessed data in the pretreatment and val-
idation of forest fire prediction data:
Table 3: Numerical Results of the Comparison of Forest
Fire Prediction Values.
Model Accuracy F1 Score
Random Forest 63.71% 67%
Decision Tree 66.2% 67.73%
Logistic Regression 62.05% 67.61%
Na
¨
ıve Bayes 55.13% 69.89%
Multilayer Perceptron 86.70% 87.93%
The Multilayer Peceptron approach has an accu-
racy value of 86.70% and an F1 Score of 87.93%
greater than the Random Forest, Decision Tree, Lo-
gistic Regression, and Nave Bayes methods, accord-
ing to the results of analyzing forest fire prediction
data using traditional machine learning algorithms.
Figure 2 illustrates it with a value differential of be-
tween 15 and 20 percent.
Figure 2: Results of Conventional Machine Learning Algo-
rithm Values.
The test was run by optimizing the traditional ma-
chine learning algorithm using the multilayer pecep-
tron approach, which has a higher value than other
methods and related research, as evidenced by an-
alyzing the confusion matrix. The F1 score of the
multilayer perceptron is 87.93% greater than that of
the Random Forest, Decision Tree, Logistic Regres-
sion, and Nave Bayes methods. Its accuracy value is
86.70%. The average accuracy difference is 14.03%,
and the F1 score is 5.29%, according to these figures.
ICAISD 2023 - International Conference on Advanced Information Scientific Development
36
4 CONCLUSION
The preprocessing process for forest fire prediction
data research has been obtained from the forest fire
prediction research using data selection, data clean-
ing, and feature selection. Conventional machine
learning data mining approaches can process data
well with the multilayer perceptron method with the
parameters train 0.7 and test 0.3. The multilayer per-
ceptron method yields an accuracy of 86.70% and an
F1 score of 87.93% with a hidden layer size of 32.32,
which is higher than the Random Forest, Decision
Tree, Logistic Regression, and Nave Bayes methods.
This value is quite dominant compared to other meth-
ods. This research can determine the proportion of
the possibility of forest fires occurring, and it is antic-
ipated that in future research, it can be developed by
deepening the size of the hidden layer for more accu-
rate reporting of forest fires.
REFERENCES
Agustyaningrum, C., Gata, W., Nurfalah, R., and Radiyah,
U. (2020). Komparasi algoritma naive bayes , random
forest dan svm untuk memprediksi niat. J. Inform,
20(2).
Agustyaningrum, C., Haris, M., Aryanti, R., and Misriati, T.
(2021). Online shopper intention analysis using con-
ventional machine learning and deep neural network
classification algorithm. J. Penelit. Pos dan Inform,
11(1):89–100,.
Ayuningtyas, F. and Prasetyo, S. (2020). Pemanfaatan
teknologi machine learning untuk klasifikasi wilayah
risiko kekeringan di daerah istimewa yogyakarta
menggunakan citra landsat 8 operational land imager
(oli. J. Transform, 18(1):13,.
Baranovskiy, N. and Zharikova, M. (2014). A web-oriented
geoinformation system application for forest fire dan-
ger prediction in typical forests of the ukraine. Lect.
Notes Geoinf. Cartogr, 0(199669):13–22,.
Cortez, P. and Morais, A. (2007). A data mining approach to
predict forest fires using meteorological data. In Proc.
13th Port. Conf. Artif. Intell, pages 512–523,. Online].
Available:.
Husen, D., Sandi, D., and Bumbungan, S. (2022). Analisis
prediksi kebakaran hutan dengan menggunakan algo-
ritma random forest classifier kebakaran hutan dan la-
han di indonesia telah menjadi perhatian dunia inter-
nasional khususnya sejak kebakaran hutan yang ter-
jadi pada tahun 80-an [ 2 ]. penyebab kebaka.
Irfan, M., Sumbodo, B., and Candradewi, I. (2017). Sistem
klasifikasi kendaraan berbasis pengolahan citra digital
dengan metode multilayer perceptron. IJEIS (Indone-
sian J. Electron. Instrum. Syst, 7(2):139,.
Kabir, M., Ashraf, F., and Ajwad, R. (2019). Analysis of
different predicting model for online shoppers’ pur-
chase intention from empirical data. Conf. Comput.
Inf. Technol. ICCIT.
Leonardo, R., Pratama, J., and Chrisnatalis, C. (2020).
Perbandingan metode random forest dan na
¨
ıve bayes
dalam prediksi keberhasilan klien telemarketing. J.
Teknol. Dan Ilmu Komput. Prima, 3(2):1–5,.
Manalu, D., Zarlis, M., Mawengkang, H., and Sitompul, O.
(2021). Forest fire prediction in northern sumatera us-
ing support vector machine based on the fire weather
index. In no, pages 187–196,.
Negoro, N., diana, M., Ula, M., and Insani, F. (2022).
Analisis kebakaran pada hutan dan lokasi lahan di
provinsi riau menggunakan metode c4.5. Maret,
7(1):107–114,.
Nurachim, R. (2019). Pemilihan model prediksi indeks
harga saham yang dikembangkan berdasarkan algo-
ritma support vector machine ( svm ) atau multilayer
perceptron ( mlp ) studi kasus : Saham pt telekomu-
nikasi indonesia tbk.
Pham, B. (2020). Performance evaluation of machine learn-
ing methods for forest fire modeling and prediction.
Symmetry (Basel, 12(6):1–21,.
Pratiwi, T., Irsyad, M., and Kurniawan, R. (2021). Klasi-
fikasi kebakaran hutan dan lahan menggunakan algo-
ritma na
¨
ıve bayes (studi kasus: Provinsi riau. J. Sist.
dan Teknol. Inf, 9(2):101,.
Ratnawati, L. and Sulistyaningrum, D. (2019). Penera-
pan random forest untuk mengukur tingkat keparahan
penyakit. J. Sains Dan Seni Its, 8(2):71– 77,.
Saputra, R., Puspitasari, D., and Baidawi, T. (2022). De-
teksi kematangan buah melon dengan algoritma sup-
port vector machine berbasis ekstraksi fitur glcm.
Sarofi, M., Irhamah, I., and Mukarromah, A. (2020). Identi-
fikasi genre musik dengan menggunakan metode ran-
dom forest. J. Sains dan Seni ITS, 9(1):79–86,.
Shah, K., Patel, H., Sanghvi, D., and Shah, M. (2020). A
comparative analysis of logistic regression, random
forest and knn models for the text classification. Aug-
ment. Hum. Res, 5(1).
Yandi, J., Kurniawan, T., Negara, E., and Akbar, M. (2021).
Prediksi lokasi titik panas kebaran hutan menggu-
nakan model regresion svm (support vector machine)
pada data kebakaran hutan daops manggala agni oki
provinsi sumatera selatan tahun 2019. InfoTekJar J.
Nas. Inform. dan Teknol. Jar, 6(1):10–15,.
Zulfikar, W. and Lukman, N. (2016). Perbandingan naive
bayes classifier dengan nearest neighbor untuk identi-
fikasi penyakit mata. J. Online Inform, 1(2):82–86,.
Forest Fire Data Analysis Using Conventional Machine Learning Algorithms
37