SUPPORT VECTOR MACHINE-BASED FIRE OUTBREAK DETECTION SYSTEM

International Journal on Soft Computing, Artificial Intelligence and Applications (IJSCAI), Vol.8, No.2, May 2019
DOI :10.5121/ijscai.2019.8201 1
SUPPORT VECTOR MACHINE-BASED FIRE
OUTBREAK DETECTION SYSTEM
Uduak Umoh, Edward Udo and Nyoho Emmanuel
University of Uyo, Faculty of Science, Department of Computer Science, Uyo, Akwa
Ibom State, Nigeria, PMB 1017
ABSTRACT
This study employed Support Vector Machine (SVM) in the classification and prediction of fire outbreak
based on fire outbreak dataset captured from the Fire Outbreak Data Capture Device (FODCD). The fire
outbreak data capture device (FODCD) used was developed to capture environmental parameters values
used in this work. The FODCD device comprised DHT11 temperature sensor, MQ-2 smoke sensor, LM393
Flame sensor, and ESP8266 Wi-Fi module, connected to Arduino nano v3.0.board. 700 data point were
captured using the FODCD device, with 60% of the dataset used for training while 20% was used for
testing and validation respectively. The SVM model was evaluated using the True Positive Rate (TPR),
False Positive Rate (FPR), Accuracy, Error Rate (ER), Precision, and Recall performance metrics. The
performance results show that the SVM algorithm can predict cases of fire outbreak with an accuracy of
80% and a minimal error rate of 0.2%. This system was able to predict cases of fire outbreak with a higher
degree of accuracy. It is indicated that the use of sensors to capture real world dataset, and machine
learning algorithm such as support vector machine gives a better result to the problem of fire management.
KEYWORDS
Support Vector Machine, Fire Outbreak, Environmental Parameters, Sensors
1. INTRODUCTION
Fire is a combustion process which results in light, smoke, heat, flame and various hazardous
gases [1]. Heat, oxygen and fuel are the three major elements of fire. The nature of the fire
depends on the proportion of each of these elements [2]. Even though fire has always been useful
for promoting the development of human society and it is extensively used in lot of applications
such as cooking, manufacturing process and other processes, when fire is out of control, i.e., fire
hazard or fire outbreak, it can cause a serious effect to human life as well as property. Fire
outbreak can also damage the ecological environment on a great level. Fire outbreak detection is
the process of sensing of one or more phenomena resulting from fire such as smoke, heat, infrared
light radiation or gas. Fire outbreak detection is usually done manually by visual observation but
it is a hazardous job that can put the life of a human being in danger. Thus, in order to prevent fire
outbreak and reduce losses due to fire outbreak, and for human safety, there is a great requirement
to develop an intelligent system for fire outbreak detection. By putting this type of system to
perform the fire outbreak detection task in fire-prone areas of a building, it can help in avoiding
fire accidents and the loss of lives. Traditionally the fire detection is done using only smoke
detectors which is less reliable technique and results in false alarms and these fire detectors are
unable to respond quickly and in reliable manner in hazardous conditions [3]. Unlike the
traditional fire outbreak detection devices, this system implements a sensor approach in predicting
the fire outbreak. This system is based on a machine learning technique called support vector
machine for the detection of fire outbreak.

2
Support vector machines are data-driven and nonlinear approaches used for pattern classification
problems which do not incorporate problem domain knowledge [4] [5]. Support vector machines
are supervised learning techniques, which are a relatively new class of learning machines [6].
Furthermore, an SVM not only effectively learns complex relationships without heuristic feature
parameters but is also suited to the use of a limited trained data set. The main goal of SVM theory
is to use kernel functions to map the training data into high-dimensional feature space. A hyper
plane is then found in this feature space that maximizes the margin between categories. The
constructed hyper plane can then be used as a basis for classifying vectors of unknown
classification with the largest margin between classes.
2. RELATED WORKS
In [7]), a Fire detection based on vision sensor and support vector machines is proposed, a new
vision sensor-based fire-detection method for an early-warning fire-monitoring system. First,
candidate fire regions are detected using modified versions of the detection of moving regions and
fire-colored pixels methods. Next, since fire regions generally have a higher luminance contrast
than neighbouring regions, a luminance map is made and used to remove non-fire pixels.
Thereafter, a temporal fire model with wavelet coefficients is created and applied to a two-class
support vector machines (SVM) classifier with a radial basis function (RBF) kernel. The SVM
classifier is then used for the final fire-pixel verification. Experimental results showed that the
proposed approach was more robust to noise, such as smoke, and subtle differences between
consecutive frames when compared with the other method.
In [8] an SVM based forest fire detection using static and dynamic features is proposed. It is an
approach for automatic forest fire detection from video based on 3D point cloud of the collected
sample fire pixels, Gaussian mixture model is built and helps segment some possible flame
regions in single image. Then the new specific flame pattern is defined for forest, and three types
of fire colours are labelled accordingly. With 11 static features including colour distributions,
texture parameters and shape roundness, the static SVM classifier is trained and filters the
segmented results. Using defined overlapping degree and varying degree, the remained candidate
regions are matched among consecutive frames. Subsequently the variations of colour, texture,
and roundness, and area, contour are computed, and then the average and the mean square
deviation of them are obtained. Together with the flickering frequency from temporal wavelet
based Fourier descriptors analysis of flame contour, 27 dynamic features are used to train the
dynamic SVM classifier, which is applied for final decision. The approach was tested with dozens
of video clips, and it can detect forest fire while also recognizing the fire like objects, such as red
house, bright light and flying flag.
In [9] an intelligent fire detection and mitigation system is studied. The study proposed an
adaptive fusion method for fire detection based on three sensors - Flame sensor, Temperature
sensor and Gas sensor. Arduino Uno with Flame sensor, Temperature sensor and Gas sensor are
used as source of the input data. Arduino IDE was used for system implementation. Sometimes
the actual fire can be taken as the deceptive fire because the system considered multiple numbers
of sensors’s input to determine the actual fire event. [10] On “Fire Detection Using Support
Vector Machines (SVM)” proposes a robust approach to wildfire detection using support vector
machine (SVM) algorithm. The result was compared with logistic regression. SVM proofs
superior in detecting cases of wildfire.
A design and implementation of a fire and obstacle detection control system is investigated based
on Arduino Uno, using fire sensor, temperature sensor, smoke sensor, and PIR motion sensor on
automobile [11]. The system was implemented with Matlab and used for system development.

3
The microcontroller unit responds to the instructions sent by the Matlab R2014 software
according to the necessity of the application as well as triggers the relay and fan upon critical
situations.
Design of a fire detection algorithm based on Fuzzy Logic, using temperature sensor to estimate
the direction of fire disaster is carried out [12] While the sensor nodes used has the ability to
periodically collect ambient temperature, communicate with neighbor sensors, and store the
neighbors’ information such as Nodes’ ID, and the Nodes’ coordinate, Temperature sensor is not
enough to detect fire outbreak. Sensor node could not know their exact location in the grid. [13]
presents an intelligent Fire Monitoring and Warning System (FMWS) that is based on Fuzzy
Logic to identify the true existence of dangerous fire and send alert to Fire Management System
(FMS). The study discusses design and application of a Fuzzy Logic Fire Monitoring and
Warning System that also sends an alert message using Global System for Mobile
Communication (GSM) technology. One of the most interesting real world phenomena that can
be monitored by WSN is indoor or outdoor fire.
In [14] a study explores two fuzzy logic approaches, with temporal characteristics, for monitoring
and determining confidence of fire in order to optimize and reduce the number of rules that have
to be checked to make the correct decisions. The study assumes that this reduction may lower
sensor activities without relevant impact on quality of operation and extend battery life directly
contributing the efficiency, robustness and cost effectiveness of sensing network. [15],
implements fuzzy logic on the information collected using sensors by passing them on to the
cluster head using event detection mechanism. Multiple sensors are used for detecting probability
of fire as well as direction of fire. Each sensor node consists of multiple sensors that will sense
temperature, humidity, light and CO density for calculating probability of fire and azimuth angle
for calculating the direction of fire. [16], employs fuzzy logic in fire detection system, which is
also multi sensor based. The driving force behind this was to develop an efficient but a simple
and error free system for critical applications. [17], presents an approach for fire detection and
estimation robots based on type-2 fuzzy logic system that utilizes measured temperature and light
intensity to detect fires of various intensities at different distances.
In [18], experiment is performed using the prototype, fire detection devices with an IoT
connection to speed up the monitoring of fire hotspots. The use of fuzzy logic minimises false
warnings from fire detection devices. The prototype can be used as a medium of learning for high
school students majoring in computer engineering and networking. [19], develops a WiFi
ESP8266 Board Compatible with Arduino IDE. [20], study temperature measurement with a
Thermistor and an Arduino. [21], develop an IoT-based intelligent modeling of smart home
environment for fire prevention and safety [22], carried out forest fire detection using wireless
sensor. [23] [24] carry out a review of wildfire detection using social media. [25], designs and
implement forest-fires surveillance system based on wireless sensor networks for South Korea
Mountains.
3. RESEARCH METHODOLOGY
In this paper, SVM-based fire outbreak detection system is developed and is applied to detect
cases of fire outbreak via the use of temperature sensor, smoke sensor and flame sensor. These
sensors are connected to an Arduino nano microcontroller unit. The Arduino microcontroller
interfaces with the system peripherals (sensors) through the input/output (I/O) pins. Values sensed
by these sensors are read by the Arduino board via a C program running on the Arduino’s
ATMega processor. The read parameter values are sent in a request parameter as a sequence of
text via an ESP8266 Wi-Fi module to a web interface that captures, pre-processes, stores the data

4
for use by the system. The stored dataset is visualized in order to descend their relationship. 700
data points are captured for the dataset with the use of FODCD. 60% of the data are used for
training dataset while 20% of the dataset are employed for testing and validation. The conceptual
architecture of the SVM-based fire outbreak detection system is presented in Figure 1.
3.1. ARCHITECTURE OF THE SYSTEM
Standard The conceptual architecture of the proposed system depicts the structure, components,
and the relationship between each component of the system. The conceptual architecture
comprises the following: (i) Sensors (ii) Arduino Microcontroller (iii) Wifi Module (iv) Data
Receptor (v) Dataset (vi) Classifier (vii) Data Visualization. Three SVM-based fire outbreak
detection input sensors are employed in this work for environmental data capturing as shown in
Figure 2(a, b, c) respectively. They include – DHT11 Temperature sensor for temperature
measurement, MQ-2 Smoke sensor for measurement of the amount of smoke gas in the air, and
LM393 flame sensor to measure light intensity. These sensors provide a source of input values to
the system where environmental parameters related to fire outbreak are captured. The study use
Arduino Nano V3.0 Microcontroller Board with micro chip, adapted from [18] and presented in
Figure 3, to interface all the system peripherals (sensors, and WI-FI module). The Arduino
Microcontroller board serves as the brain of the fire outbreak data capture device. Its main
function is to read parameter values from the connected sensors and send same to the classifier
module via an ESP8266 WIFI module. The ESP8266 is used to add wireless network
functionality to the fire outbreak data capture device (FODCD). The ESP8266 acts as a web
server; hence its data can be viewed using a web interface programmed as an android application.
The ESP8266 WIFI module used in this work is presented in Figure 4.
Figure1. The Conceptual Architecture of SVM-Based Fire Outbreak Detection System
Classifier
Model
Predicted
Result
Report
DB
Data Visualization
Classification
SVM
Test Data
Micro Controller
Unit (MCU)
Wifi Module
(ESP8266)
Binary Data
System Sensors
Flame Sensor
Smoke Sensor
Temperature
Sensor
Data Receptor
Data
Pre-processor
User Interface
Dataset

5
Figure 2: (a) MQ-2 Smoke Sensor (b) DHT11 Temperature Sensor (c) LM393 Flame Sensor employed in
SVM-Based Fire Outbreak Detection
Figure 3. Arduino Nano V3.0 board for Interfacing Sensors and WI-FI module
Figure 4: ESP8266 WIFI Module
The Data receptor is implemented as a software module. It sends an HTTP request for data to the
ESP8266 WIFI module and accepts a response in the form of “HTTP Response Text”. The Data
Receptor reads and extracts the relevant parameters (i.e. temperature, smoke, and flame values)
from the “HTTP Response Text” using the heuristic algorithm in Figure 5. Figure 6 gives the
circuit diagram of the fire outbreak data capture device showing the hardware components, their
pin connection, and the relationship of each component with the other as used in this paper.
(a) (b)
(c)

6
Figure 5. Heuristic algorithm for data reading and extraction of Input Parameter Values by Data Receptor
algorithm Pre-process(msg):Arr[]
begin:
Define temp, smoke, flame;
Define dataSize N;
Define indexOfAsteric indx1, indexOfAsh indx2;
Define substringVariablesubStr;
Define array Variable Arr[];
//find the position of char *
Loop for i = 0 to N
Let x = charAt(i)
If x == “*”
indx1 = i;
break;
end if
end for
//find the index of char #
Loop for j = 0 to N
Let y = charAt(j)
if y == “#”
Indx2 = j;
break;
end if
end for
//extract data substring and split to an array
if( indx1 > -1 AND indx2 > -1)
subStr = subString(indx1, indx2);
Arr[] = split(subStr, “,”);
return Arr[];
end if
end

7
Figure 6. Circuit diagram of the fire outbreak data capture device
3.2.THE PROTOTYPE OF FIRE OUTBREAK DATA CAPTURE DEVICE
The FODCD device is developed for the purpose of extracting environmental parameters values
related to fire outbreak. The FODCD provides the input parameter values used in testing the
performance of this system. The breadboard is designed to connect the necessary components of
the device together to help in testing and debugging of the device before the final soldering is
done. The prototype of the fire outbreak data capture device is presented in Figure 7;
Figure 8. Prototype of the fire outbreak data capture device
LM393
VDDAO DOGND
MQ-2 VIN
AO
DO
GND
ESP8266
Wifi
Module
Arduino
MCU
Power
Voltage
Regulator
3.3v
DHT11
VDDNUL
L
DAT
A
GN
D
BAT

8
3.3.FIRE OUTBREAK DATASET
The dataset used in this work is captured with the FODCD. The data collected includes the
temperature, smoke, and flame parameter values. The dataset is made up of three features
(temperature, smoke and flame) and one output label called “Fire Outbreak Detection”. The label
with values 0 is the negative class (i.e. “NoFireOutbreak”) while 1 is the positive class (i.e.
“FireOutbreak”). The sample of this data is presented in Table 1.
Table1. Fire Outbreak Collected Dataset
Temp Smoke Flame Label
25.977 50.529 464.37 0
36.301 60.853 464.47 0
18.554 36.632 359.63 0
29.317 52.735 387.89 0
16.351 30.961 234.26 0
19.748 38.247 283.3 0
25.229 47.722 471.83 0
33.063 54.949 593.49 0
34.735 66.18 574.92 0
25.041 42.024 412.61 0
15.946 27.564 208.08 0
28.005 55.083 450.8 0
42.141 74.721 796.43 0
17.825 30.052 336.47 0
17.423 32.546 312.48 0
15.365 28.35 211.6 0
30.165 56.058 558.45 0
42.137 81.57 753.23 1
33.566 65.797 483.05 0
37.682 66.021 516.66 0
45.647 79.257 920.62 1
15.358 27.124 277.24 0
19.993 35.956 301.45 0
35.196 57.753 604.37 0
27.631 47.253 409.88 0
41.154 73.372 666.96 0
32.217 53.757 528.32 0
45.072 77.857 848.01 1
19.699 31.73 323.23 0
24.379 44.86 311.9 0
16.406 29.229 257.16 0
48.478 94.916 826.49 1
48.24 88.208 845.21 1
16.552 30.01 218.67 0
38.36 64.808 503.67 0
39.138 70.889 791.07 0
45.98 84.589 822.88 1
49.955 86.207 872.83 1
24.913 46.242 374.64 0
36.601 58.797 621.71 0
32.489 56.149 572.02 0
15.163 30.077 221.83 0
34.102 62.306 627.75 0
32.325 55.507 410.25 0
18.398 34.357 329.47 0
45.018 73.837 757.65 1
24.17 43.099 494.4 0
28.059 49.878 421.45 0
46.949 77.574 610.23 1
43.309 82.043 575.21 1
39.149 64.331 663.74 0
20.091 33.917 326.08 0
15.362 28.662 305.08 0
34.936 67.158 455.75 0
33.721 66.066 595.37 0
16.054 30.052 315.77 0
20.264 37.739 376.55 0
45.774 90.27 670 1

9
3.4.FIRE OUTBREAK DATA VISUALIZATION
Data visualization are performed on the fire outbreak captured data where different graphs are
plotted to visualize the collected dataset in order to get a better understanding of the data
structure. The different tools used in the dataset visualization include Lag plot, Correlation graph,
and Andrew curveline. The results are obtained based on each visualization tool and are displayed
in section four of this paper.
3.5.SUPPORT VECTOR MACHINE FOR THE FIRE OUTBREAK DETECTION
Support Vector Machine (SVM) is used to classify the data points (temperature, smoke and
flame) into a positive class “FireOutbreak” with label 1 or negative class “NoFireOutbreak” with
label 0. The Input data (training data) are formulated as shown in Equation 1;
(𝑥1, 𝑦1) . . . . . . . (𝑥 𝑛, 𝑦𝑛) (1)
Where:
𝑥 is the feature set, 𝑦 is the label, 𝑥𝑖 = 𝑥𝑖
1
, 𝑥𝑖
2
, . . . . . . 𝑥𝑖
𝑑
, Where:𝑥𝑖
𝑗
: is a real value and 𝑦𝑖 = {0, 1}
with 0 representing “No Fire Outbreak” and 1 representing “Fire Outbreak”
In this work the RBF (Radial Basis Function) Kernel is used to map the non-separable training
data from input space to feature space in order to find an optimized hyper plane that correctly
segregates the data. The RBF kernel is presented in Equations 2 and 3 respectively.
𝐾(𝑥𝑖⃗⃗⃗ , 𝑥𝑗⃗⃗⃗ ) = ∅(𝑥𝑖⃗⃗⃗ ) 𝑇
∅(𝑥𝑖⃗⃗⃗ ) (2)
𝐾(𝑥𝑖⃗⃗⃗ , 𝑥𝑗⃗⃗⃗ ) = exp(−𝛾 ||𝑥𝑖 − 𝑥𝑗||2
) (3)
Where: 𝛾 is
1
2𝜎2 > 0, 𝑥𝑖 is the support vector points, 𝑥𝑗 is the feature vector points in the
transformed space and 𝐾(𝑥𝑖⃗⃗⃗ , 𝑥𝑗⃗⃗⃗ ) is the Kernel function. The kernel function calculates the dot
product of the mapped data points in the transformed feature space. The optimal hyper plane that
segregates between the two classes (“FireOutbreak”, “NoFireOutbreak”) is found using equation
4;
𝑤 𝑇
. 𝑥 + 𝑏 = ∑ 𝛼𝑖 𝑦𝑖 ∅(𝑥𝑖⃗⃗⃗ ) 𝑇
∅(𝑥𝑖⃗⃗⃗ ) + 𝑏𝑙
𝑖=1 = 0 (4)
The classification frontiers are found using;
𝑤 ∅(𝑥) + 𝑏 = 1: for points labelled as “FireOutbreak” (5)
𝑤 ∅(𝑥) + 𝑏 = 0: for points labelled as “NoFireOutbreak” (6)
The optimal weight vector (w) is given by;
𝑤⃗⃗ = ∑ 𝛼𝑖 𝑦𝑖 ∅(𝑥𝑖⃗⃗⃗ )𝑙
𝑖=1 (7)
This work uses the dual formulation of SVM algorithm presented as a maximization problem
over 𝛼 is given in Equation 8;

10
max ∑ 𝛼𝑖 −
1
2
∑ ∑ 𝛼𝑖
𝑙
𝑗=1 𝛼𝑗 𝑦𝑖
𝑙
𝑖=1 𝑦𝑗 ∅(𝑥𝑖⃗⃗⃗ ) 𝑇
∅(𝑥𝑖⃗⃗⃗ )𝑙
𝑖=1 (8)
Subject to:
0 < 𝛼 < 𝐶 and ∑ 𝛼𝑖
𝑙
𝑖=1 𝑦𝑖 = 0
Where: 𝛼𝑖: is the weight vector, 𝑦: is the label vector, ∅(𝑥𝑖⃗⃗⃗ ) 𝑇
∅(𝑥𝑖⃗⃗⃗ ): is the kernel function, 𝐶: is
the intercept. The decision function g used in making prediction is given as;
𝑔(𝑥) = 𝑠𝑔𝑛(𝑤⃗⃗ 𝑇
𝑥 + 𝑏) ⇒ 𝑠𝑔𝑛(∑ 𝛼𝑖 𝑦𝑖 ∅(𝑥𝑖⃗⃗⃗ ) 𝑇
∅(𝑥𝑖⃗⃗⃗ ) + 𝑏𝑙
𝑖=1 ) (9)
Where: 𝑔(𝑥): is the predicted label, 𝑠𝑔𝑛: is the sign of (𝑤⃗⃗ 𝑇
𝑥 + 𝑏) (i.e. 1 or 0), 𝛼𝑖: is the weight
vector. The general steps taken to accomplish the classification and prediction of fire outbreak are
given in Figure 9.
Figure 9: General Steps for Classification and Prediction of Fire Outbreak
3.3. PERFORMANCE METRICS FOR FIRE OUTBREAK DETECTION
The True Positive Rate (TPR), False Positive Rate (FPR), Precision, Error rate, and accuracy
performance metrics are employed to evaluate the accuracy of the developed fire outbreak
system. The models are presented as follows;
𝑇𝑃𝑅 (𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 𝑜𝑟 𝑅𝑒𝑐𝑎𝑙𝑙) =
𝑇𝑃
𝑇𝑃+𝐹𝑁
(10)
𝐹𝑃𝑅 =
𝐹𝑃
𝐹𝑃+𝑇𝑁
(11)
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑇𝑃
𝑇𝑃+𝐹𝑃
(12)
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃+𝑇𝑁
𝑁
(13)
𝐸𝑟𝑟𝑜𝑟 𝑅𝑎𝑡𝑒 =
𝐹𝑃+𝐹𝑁
𝑁
(14).
4. RESULT AND DISCUSSION
This study employed Support Vector Machine (SVM) in the classification and prediction of fire
outbreak based on fire outbreak dataset captured from the FODCD. The fire outbreak data capture
device (FODCD) used was developed to capture environmental parameters in this work. The
FODCD device comprised DHT11 temperature sensor, MQ-2 smoke sensor, LM393 Flame
sensor, and ESP8266 Wi-Fi module, connected to Arduino nano v3.0.board. 700 data point were
STEP 1: Load input training data
STEP 2: Separate data into 𝑥𝑖(feature set) and 𝑦𝑖(labels)
STEP 3: Map data from input space to feature space using
RBF Kernel
STEP 4: Find an optimal hyper plane using Equation (6)
STEP 5: Find classification frontiers (support vectors)
STEP 6: Learn by maximizing 𝛼 in Equation (8)
STEP 7: Make prediction using Equation (9)

11
captured using the FODCD device, with 60% of the dataset used for training while 20% was used
for testing and validation respectively. The SVM model was evaluated using the True Positive
Rate (TPR), False Positive Rate (FPR), Accuracy, Error Rate (ER), Precision, and Recall
performance metrics. Figure 10 shows the FODCD developed in this paper and used for capturing
the input parameter (temperature, smoke and flame) values as shown in Table 1. Figure 11 gives
the line graph indicating the levels of temperature values from Table 1. From the graph it is
shown that the higher the line in the graph, the higher the value of temperature. Figure 12 presents
the filled line graph showing the levels of the different smoke values received from the FODCD
and used in training and testing of the system. The flame data is collected using FODCD and the
graph of the result showing the different flame values is presented in Figure 13.
Figure 14 presents the Lag plot of the captured fire outbreak dataset used to check the
randomness, suitability, outliers of the values in the dataset. Data is said to be random when there
is no visible pattern in the lag plot. From Figure 14 (lag plot), we can observe a pattern in the plot
which implies that our dataset is not random, hence the dataset is good for the machine learning
task at hand. Figure 15 is the correlation graph used to visualize the relationship among the
features and the relationship between the fire outbreak features and the labels (output variables).
The correlation ranges from -1 to 1, with -1 meaning that the dataset is highly negatively
correlated and 1 meaning that the dataset is highly positively correlated. From the Figure 15, the
correlation between our features and the labels is about 0.75. This implies that the features are
highly correlated to the predictor (class label). Figure 16 presents the Andrews curve used in
visualizing multidimensional data by mapping each observation onto a function. The Andrew
curve for the fire outbreak dataset shows the closeness of data points from one class to that of
other class. In the plot, each colour used, represents a class (0 for negative class i.e.
“NoFireOutbreak”, 1 for positive class i.e. “FireOutbreak”) and we can easily note the lines
representing samples from the same class exhibit similar pattern. Figure 16 shows that there is a
significant level of uniqueness between points that belong to the negative class (“no fire
outbreak”) and points that belong to the positive class (“fire outbreak”).
Figure 17 gives the training result from the support vector machine for fire outbreak detection. It
shows the segregation of the data points into their classes. The red points labeled 0 represents the
negative class called “No Fire Outbreak” and the blue points labeled 1 represents the positive
class called “Fire outbreak”. The Circles are the support vectors (i.e. data points that participate in
the deciding the segregating hyper plane). The figure shows a proper segregation of the data
points into its various classes (“FireOutbreak” and “NoFireOutbreak”). Figure 18 shows the graph
of correctly classified points during model testing. The blue points are the true positives (i.e.
points that are truly predicted as “FireOutbreak”) while the red points in the graph are true
negative points (i.e. points that are truly predicted as “NoFireOutbreak”). Figure 19 shows the
misclassified points. The red points in Figure 19 are points that are false negative points (i.e.
points that are wrongly classified as “NoFireOutbreak”).
Figure: 10. Fire Outbreak Data Capture Device

12
Figure 11. The Graph of Temperature Data Captured for Fire outbreak Detection
Figure12. The graph of the Captured Smoke Data for Fire outbreak Detection
Figure 13. Graph of the Values of Flame Data Captured

13
Figure 14. Lag Plot of the Fire Outbreak Dataset
Figure 15. The graph of Correlation between the Fire the Outbreak Input and Output Variable
Figure 16. The Andrew Curve of the Fire Outbreak Dataset

14
Figure 17. SVM Training result for Fire Outbreak Detection
Figure 18. Correctly Classified Points
Figure 19. Misclassified points
The accuracy of the test result is evaluated using true positive, true negative, false positive, false
negative, true positive rate, false positive rate, accuracy, precision and error rate. The test
performance results of the model are presented in Table 2. From Table it is observed that, the
values in the matrix represents true negative (i.e. 111), true positive (i.e. 1), false negative (i.e.
28), and false positive (i.e. 0). Figure 20 shows the Receiver Operating Characteristic (ROC)
curve indicating the accuracy of the model on the test data. The accuracy rises when the curve is

15
closer to the top left corner and deteriorates when the curve moves otherwise. From the ROC
curve in Figure 20, the accuracy of 0.8 is attained from the model. The confusion matrix of the
model testing is presented in Figure 21.
Table 2: Test Performance Result
Metric Symbol Value
True Positive TP 1
True Negative TN 111
False Positive FP 0
False Negative FN 28
True Positive Rate (Recall) TPR 0.034483
False Positive Rate FPR 0
Accuracy - 0.8 ≈ 80%
Precision - 1
Error Rate - 0.2
Figure 20. Test accuracy
Figure 21. Confusion Matrix for testing result
5. CONCLUSION
This study employed Support Vector Machine (SVM) in the classification and prediction of fire
outbreak based on fire outbreak dataset captured using Fire Outbreak Data Capture Device
(FODCD). The fire outbreak data capture device (FODCD) is developed to capture environmental
parameters values used in this work. The FODCD device comprised DHT11 temperature sensor,
MQ-2 smoke sensor, LM393 Flame sensor, and ESP8266 Wi-Fi module, connected to Arduino
nano v3.0.board. 700 data point were captured using the FODCD device, with 60% of the dataset

16
used for training while 20% was used for testing and validation respectively. The results indicate
that the support vector machine-based fire outbreak detection has been able to proffer solution to
the problems associated with fire outbreak by providing continuous monitoring of environmental
changes that are responsible for fire outbreak such as Temperature, Smoke and Flame. The
system is also able to detect cases of fire outbreak by learning the parameters that could lead to
fire outbreak. The system can predict a case of fire outbreak with accuracy of 80% with a minimal
error rate of 0.2%. It is possible to detect and take immediate action during fire outbreak. When
this is done, we can protect lives and properties as well as reduce the rate of depression resulting
from fire disaster. The combination of sensors, and support vector machine learning algorithm
gives a better result to the problem of fire management. In the future, more input parameters can
be added and the system can be incorporated with deep learning tools for a better performance.
REFERENCES
[1] Sharma, D. (2016). Implementation of Artificial Neural Fuzzy Inference System in a Real Time Fire
Detection Mechanism. International Journal of Computer Applications, Volume 146, 31-36.
[2] Sowah, R. (2014). A Review on Forest Fire Detection Techniques. International Journal of Distributed
Sensor Networks, 2-8.
[3] Mobin, M. A., Rafi, M., Islam, M. N. (2016). An Intelligent Fire Detection and Mitigation System
Safe from Fire. International Journal of Computer Applications , Volume 133, 1-6.
[4] Wang L., and X. Fu X. (2005). Data Mining with Computational Intelligence. Springer.
Sloin A., Burshtein D. (2008). Support vector machine training for improved hidden Markov
modeling. IEEE Transactions on Signal Processing, Volume 56, 172–188.
[5] Khandoker A. H., Palaniswami M., Karmakar C., K. (2009). Support vector machines for automated
recognition of obstructive sleep apnea syndrome from ECG recordings. IEEE Transactions on
Information Technology in Biomedicine, volume. 13, no. 1, pp. 37–48.
[6] Byoungchul k., Kwang-Ho C., Jae-Yeal N. (2009). Fire detection based on vision sensor and support
vector machines. Fire Safety Journal. Volume 44, 322-32.
[7] Zhao J., Zhong Z., Han S., Zhang D. (2010). SVM based forest fire detection using static and dynamic
features. Computer Science and Information Systems Volume 8. 821-841.
[8] Hassan, M. K., Davood, N., Alireza, A., Mehdi, S. (2013). Swarm optimization tuned Mamdani fuzzy
controller for diabetes delayed model. Turkish Journal of Electrical Engineering & Computer
Sciences, 31-46.
[9] Vigneshwara SR, Shanthakumari S. S., Ranganathan V. (2017). Fire Detection using Support Vector
Machine (SVM). International Journal of Science and Research (IJSR). Volume 6.
[10] Khule, V., Jangle, N. N. (2017). Design and implementation of a fire and obstacle detection and
control system using fuzzy logic with notiﬁcation system to avoid automobile accidents. 10th
International Conference on Recent Trends in Engineering Science and Management. India: Newton's
School of Science and Technology, 307-313.
[11] Chu, X. J. (2010). Design of Fire Detection Algorithm Based on Fuzzy Logic and using Wireless
Sensors. Canada: Ottawa-Carleton Institute of Electrical and Computer Engineering (OCIECE).

17
[12] Sarwar, B. Bajwa, I. S., Ramzan, S., Ramzan, B, and Kausar, M., (2018) Design and Application of
Fuzzy Logic Based Fire Monitoring and warning Systems for Smart Buildings. Symmetry 2018, 10,
615, 1-24; doi: 10.3390/sym10110615
[13] Maksimović, M., Vujović, V., Perišić, B. and Milošević, V., (2014) Developing a fuzzy logic based
system for monitoring and early detection of residential fire based on thermistor sensors. Computer
Science and Information Systems 12(1):63–89 DOI:
[14] Khanna, V. and Cheema, R. K. (2013), Fire Detection Mechanism using Fuzzy Logic. International
Journal of Computer Applications (0975 – 8887) Volume 65– No.12, March 2013
10.2298/CSIS140330090M.
[15] Muralidharan, A. and Joseph, F (2014) Fire Detection System Using Fuzzy Logic. International
Journal of Engineering Sciences and Research Technology. 3(4), 1-4.
[16] Necsulescu, D. and Le, X., (2015), Type-2 Fuzzy Logic Sensor Fusion for Fire Detection Robots.
Proceedings of the 2 nd International Conference of Control, Dynamic Systems, and Robotics Ottawa,
Ontario, Canada, May 7 – 8, 2015 Paper No. 187.
[17] Listyorini, T. and Rahim, R. (2018), A prototype fire detection implemented using the Internet of
Things and fuzzy logic. World Transactions on Engineering and Technology Education Vol.16, No.1,
42-47, 2018
[18] Arduino. (2018, November 18). Arduino Basics.html. Retrieved April 20, 2018, from Arduino:
http://www.arduino.cc
[19] Robot-r-us, WeMos D1 R2 WiFi ESP8266 Board Compatible with Arduino IDE (2018), 18 October
2017, https://www.robot-r-us.com/wireless_wifi/wemos-d1-r2-wifi-esp8266-development-board-
compatible-with-ardui.html.
[20] Recktenwald, G.: Temperature Measurement with a Thermistor and an Arduino, (2013)
[21] Saeed, F.; Paul, A.; Rehman, A.; Hong, W.H.; Seo, H. (2018) IoT-Based Intelligent Modeling of
Smart Home Environment for Fire prevention and Safety. J. Sens. Actuator Netw., 7, 11.
[22] Singh, H. (2016) Forest Fire Detection using Wireless Sensor. Int. J. Sci. Eng. Res., 7, 554–562.
[23] Wang, S.; Liu, G. Automatic Fire Alarm and Fire Control Linkage System in Intelligent Building. In
Proceedings of the International Conference on Future Information Technology and Management
Engineering, Changzhou, China, 9–10 October 2010.
[24] Slavkovikj, V.; Verstockt, S.; Van Hoecke, S.; Van deWalle, R. (2014) Review of wildfire detection
using social media. Fire Saf. J., 68, 109–118.
[25] Son, B.; Her, Y.S.; Kim, J.G. (2006). A design and implementation of forest-fires surveillance system
based on wireless sensor networks for South Korea Mountains. Int. J. Comput. Sci. Netw. Secur, 6,
124–130.

18
Authors
Umoh, U. A. had received her Doctor of Philosophy (PhD) degree in Soft Computing
from University of Port Harcourt,Rivers State, Nigeria in the year 2012, Master’s degree
in Database Management System from University of Port
Harcourt, Rivers State, Nigeria in the year 2006 and Bachelor’s degree from University
of Uyo, Akwa Ibom State, Nigeria in 2007. She is currently working of as a Senior
Lecturer, in the University of Uyo and the Head of Department in the Department of
Computer Science. She has published several articles in her areas in reputable national
and international journals and has written some quality books in Computer discipline. Her area of interest
include; Soft Computing (Fuzzy System, Neural network, Hybrid models), Database System, Data
Communications, etc. She is a member of Nigerian Computer Society (NCS), Member, Computer
Professionals Registration Council of Nigeria (CPN), Mathematical Association of Nigeria (MAN),
Nigerian Women in Information Technology (NWIT), Universal Association of Computer and
Electronics Engineers (UACEE), etc.
Nyoho, E. E. is a passionate software developer. He graduated from University of Uyo with Masters of
Science degree in computer science. He has enormous experience in Software Development using
Electron, Java, C, C++, Python, R. and Web Development using PHP, Java EE, JavaScript, HTML, CSS,
Angular, Node.JS. He is also an android developer who has risen to the heights of publishing His
applications on the Google Play Store. Due to his performance in software development, He was
Awarded “Programmer of the year 2012” by National Association of Computer Science Students
(NACOSS). He has carried out numerous projects in several areas of computing, including – Mobile
Computing, Cryptography, Fuzzy Expert Systems, Image Enhancement, Steganography, Search, Machine
Learning, and Embedded Systems.
Udo, E. N. BSc(Uyo); MSc(Port Harcourt),Ph.D(UNIBEN) He is a lecturer in the
Department of Computer Science. He holds a Bachelor of Science degree in Computer
Science from University of Uyo, Nigeria, a Master degree in Computer Science from
University of Port Harcourt, Nigeria and a Ph.D degree from University of Benin,
Nigeria. He holds a professional certificate in Networking (CISCO) and Systems
Engineering (Microsoft). He is internationally recognized CISCO Certified Academy
Instructor. He is also certified as a Management Trainer by Centre for Management
Development (CMD). He is a member of Nigeria Computer Society (NCS), a member of Computer
Professionals Registration Council of Nigeria (CPN) and a member of IEEE (Computer section). He
earned FGN/TETFUND Award in 2008 and 2012 for M.Sc. and Ph.D. Research respectively.
The authors declare that there is no conflict of interest regarding the publication of this paper.

SUPPORT VECTOR MACHINE-BASED FIRE OUTBREAK DETECTION SYSTEM

More Related Content

What's hot (20)

Similar to SUPPORT VECTOR MACHINE-BASED FIRE OUTBREAK DETECTION SYSTEM (20)

Recently uploaded (20)

SUPPORT VECTOR MACHINE-BASED FIRE OUTBREAK DETECTION SYSTEM