Hybrid-intelligent Mobile Indoor Location using Wi-Fi Signals

Location Method using Data Mining Algorithms and Type-2 Fuzzy Logic Systems

Manuel Casta

on-Puga, Abby Salazar-Corrales, Carelia Gaxiola-Pacheco, Guillermo Licea,

Miguel Flores-Parra and Eduardo Ahumada-Tello

Autonomous University of Baja California, Tijuana, Mexico

Keywords:

Location, Wi-Fi, Situational Awareness, Fuzzy Logic, Data Mining, Smartphone.

Abstract:

Technology with situational awareness needs a lot of information of the environment to execute the correct

task at the correct moment. Location of the user is typical information to achieve the goal. This work proposes

a mobile application that enables the indoor location of smartphones using the potential infrastructure given by

Wireless Local Area Networks. This infrastructure goes beyond GPS (Global Position System) where signal is

weak or is not available for indoors. This application uses an alternative and unconventional method to indoor

location using Wi- Fi RSSI ﬁngerprinting as well as an estimation based on Type-2 fuzzy inference systems

provided by the developed framework JT2FIS. Wi-Fi Fingerprinting creates a radio map of a given area based

on the RSSI data from several access points (APs) and generates a set of RSSI data for a given zone location.

Consequently Data Mining is required for clustering the obtained set of data and generating the structure of

a Type-2 Mamdani or Takagi-Sugeno Fuzzy Inference System; thus new RSSI values are introduced to the

Type-2 Fuzzy Inference System to obtain an estimation of the user zone location.

1 INTRODUCTION

The growing technology and advanced algorithms can

generate a lot of useful information to be used to im-

prove other tasks. Also with recent advances in tech-

nology exists the possibility of connecting different

devices using wireless networks to transparently in-

teract with the environment and seamlessly with peo-

ple. The challenge of this idea is to ﬁnd ways to know

the particular context or situation in time and space

to make a decision based on it without the aware-

ness of the user. Devices require algorithms with the

ability to understand the context information so that

a given system can adapt to it and behave in a cer-

tain way. The behavior of a person frequently de-

pends on the context. Human nature enables us to

be aware of a lot of variables at a time that inﬂuence

a person to behavior in a certain way. Context aware-

ness could be a more suitable technological for the

needs of the moment. In order to understand the con-

text, it is necessary to have in mind different factors

which vary depending on the activity to be performed

such as current weather or temperature, atmospheric

pressure, lighting from one place, position of an ob-

ject, the place where a person is, among others. Some

other variables are more complex to detect such as the

mood of a person or the identity of other persons near

the target person. On other hand, there are variables

that are useful to determine more of these variables,

an example is the location of persons or objects; this

can help to detect the number of persons in a room or

if certain users are near. The proposed mobile appli-

cation estimates the indoor location of the user using a

hybrid-intelligent method that makes use of Wi-Fi Re-

ceived Strength Signal Indicators (RSSIs). The data

mining and fuzzy logic methods were implemented

using the developed JT2FIS framework.

1.1 Situational Awareness

In the past, objects that could understand their envi-

ronment, communicate and properly respond without

the need of direct intervention of man looked distant

due to the complexity involved. However, nowadays

this is more possible mainly by two factors: because

in the last years the technological advancement has

been increasing, and also because of the great increase

of devices connected to different networks, such as

Internet (Sanchez et al., 2014). The context is any

information that can be used to characterize the situa-

tion of an entity. An entity is a person, place, or object

that is considered relevant to the interaction between

609

Castañón-Puga M., Salazar-Corrales A., Gaxiola-Pacheco C., Licea G., Flores-Parra M. and Ahumada-Tello E..

Hybrid-Intelligent Mobile Indoor Location Using Wi-Fi Signals - Location Method Using Data Mining Algorithms and Type-2 Fuzzy Logic Systems.

DOI: 10.5220/0005369806090615

In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 609-615

ISBN: 978-989-758-096-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

a user and an application, including the user and ap-

plications themselves (Dey, 2001). Devices with the

capability of understanding the context are an impor-

tant key for the non-human intervention in technology

tasks. The context or situational awareness parameter

of this work is based on the location of the entity.

1.2 Technologies

Actual location techniques commonly use GPS, Blue-

tooth, or Radio Frequency Identiﬁcation (RFID) tech-

nologies, among others (Chai et al., 2011). Despite

that, a disadvantage of GPS is that the satellite signals

are blocked by obstacles such as walls, in addition,

variations in weather or the presence of buildings re-

sults in approximations with errors of meters, so it is

not possible to use this system as a method for indoor

location (Navarro et al., 2011)(Hwang and Donghui,

2012). On the other hand, Bluetooth technology has

limited coverage; this communication is focused on

very short distances to achieve the location. Finally

RF is an expensive solution since it involves the in-

stallation of different sensors in the area where you

will estimate the location, so it is not an economically

viable method (Navarro et al., 2011). Consequently

acceptable alternative techniques are required to ﬁt

with established infrastructures and satisfy the func-

tionality of indoor location with ease of use and an

affordable cost, as the use of technologies based on

Wi-Fi (Chai et al., 2011).

1.3 Location Methods

Triangulation and Trilateration: These methods map

RSSI as a function of distance that requires a steep

linear characterization curve in order to be properly

implemented. Functions describing these curves are

then used with live RSSI values as input to generate

an (x,y) location prediction (Chai et al., 2011). The

disadvantages of these methods are: to carry out the

synchronization of these values and that it needs a

model to determine the distance according to RSSI

values. This work is an alternative solution for local-

ization using a radio map of a given area based on

Wi-Fi RSSI data from three (or more) Access Points.

Some advantages are that this method works on in-

door environments with acceptable coverage, a min-

imum or null modiﬁcation of the area infrastructure

is required, and is a less expensive option than others

(Navarro et al., 2011). The method proposed is based

on a FIS (Fuzzy Inference System) using a cluster-

ing method (Chiu, 1994), that is a fast, one-pass al-

gorithm for estimating the number of clusters and the

cluster centers in a set of data.

2 DATA MINING

The number of generated data is increasing alongside

the increase of technology, networks, and sensors. In

recent years, the use of new information technologies

has come to help handling large amount of data. One

evolution of these technologies is the data mining ex-

traction that allows representing knowledge of data

store implicitly in big databases. Data mining con-

tributes to understand data and identify patterns, re-

lationships, and dependencies that affect the ﬁnal re-

sults. It creates predictive models that allow undis-

covered relationships through the data mining pro-

cess which are expressed as possible business rules

(Crows, 1999). Data mining is a multidisciplinary

ﬁeld that combines techniques from machine learn-

ing, pattern recognition, statistics, database, and vi-

sualization, to direct it to the extraction and inter-

pretation of a huge database. The data mining fo-

cuses on ﬁlling the need to discover, predict, and fore-

cast the possible actions with some conﬁdence factor

for each prediction (Han and Kamber, 1998). More-

over, it helps to make tactical and strategic decisions,

provided the decision power users, is able to mea-

sure actions and results in the best way, generates de-

scriptive models to explore and understand the data,

and identify patterns, relationships, and dependencies

that affect the ﬁnal results. Creates predictive models

that allow undiscovered relationships through the data

mining process which are expressed as business rules

possible (Crows, 1999). Clustering of numerical data

forms is a type of data mining. The aim of clustering

methods is to identify natural grouping of data from

a large data set, such that a concise representation of

the systems behaviour is produced (Ren et al., 2006).

Once the clusters of a data set are identiﬁed the sys-

tem behaviour can be translated to the rules of a FIS.

Each cluster is translated as one rule of the FIS. Gen-

erally more rules describe in more detail the system

behavior so a better approximation on the evaluation

can be achieved or a better accuracy is gained (Ying

et al., 1998). It is important to consider the computa-

tional cost and the robustness of the system to deﬁne

the minimal resources needed to have an acceptable

system evaluation. Some cluster features, as size and

number, are controlled by the speciﬁc parameters in-

volved in different clustering techniques.

2.1 Fuzzy C-Means

Fuzzy C-Means clustering algorithm (FCM) (Bozkir

and Sezer, 2013)(Bezdek et al., 1984) makes use of a

membership function and centroid computation pro-

cedure iteratively to ﬁnd the best centroid. The FCM

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

610

is one of the popular clustering algorithms. The ef-

fectiveness of the clustering method relies on the dis-

tance measure. FCM is the result of combining the

c-means approach with the handling of fuzzy data.

The result of this combination is adequate because it

considers the uncertainty presented in the data, avoid-

ing incorrect results, and creating crisp partitions in a

correct way (Bozkir and Sezer, 2013). Additionally

the FCM is used to acquire the adequate levels of the

set clustering parameters (Yin et al., 2013). In this

work the clustering algorithm FCM is used to obtain

the fuzzy rules of a Mamdani inference system that is

used to estimate the location of a device (smartphone

with Wi-Fi) as described in section 3.

2.2 Subtractive Method

Subtractive clustering operates by ﬁnding the optimal

data point to deﬁne a cluster center based on the den-

sity of surrounding data points. It reduces the com-

putational complexities and gives better distribution

of cluster centers in comparison with other clustering

algorithms (Ren et al., 2006). This method considers

each point as a potential center and, based on math-

ematical approximations, calculates the best choice

of center. Each cluster center can be considered as

a fuzzy rule of the system, and the cluster identi-

ﬁed represents the antecedent of this rule. The mea-

sure of potential for a data is estimated based on the

distance of this data point from all other data points

(Vaidehi et al., 2008). The identiﬁcation of a Takagi-

Sugeno (TSK) FLS using clustering involves forma-

tion of clusters in the data space and translation of

these clusters into TSK rules such that the model ob-

tained is close to the system to be identiﬁed.

3 TYPE-2 FUZZY LOGIC

The concept of a Type-2 fuzzy set was introduced by

Zadeh (Zadeh, 1965) as an extension of the concept

of usually Type-1 fuzzy sets. A Type-2 fuzzy set is

characterized by a membership function whose mem-

bership value for each element of the universe is a

membership function within the range [0, 1], unlike

the Type-1 fuzzy sets where the value of membership

is a numeric value in the range [0, 1]. The creation

of a fuzzy set depends on two aspects: the identiﬁca-

tion of a universe of appropriate values and specify-

ing a membership function properly. The choice of

membership function is a subjective process, mean-

ing that different people can reach different conclu-

sions on the same concept. This subjectivity derives

from individual differences in the perception and ex-

pression of abstract concepts and it has little to do

with randomness. Therefore, subjectivity and ran-

domness of a fuzzy set are the main difference be-

tween the study of fuzzy sets and probability theory

(Jang et al., 1997). A Type-2 Fuzzy Logic System

(FLS) can be used when it is not possible to deter-

mine exact membership grades or its uncertainty in

the rules (Cai et al., 2007). In Type-1 fuzzy sets, once

the membership function is deﬁned for a concept, this

is based on the subjective opinion of one or more in-

dividuals and shows no more than one value for each

element of the universe. In doing so, it loses some

of the ambiguity of the discussed concepts, especially

where people may have a slightly different opinion,

but they are considered valid. The Type-2 fuzzy sets

allow handling linguistic and numerical uncertainties.

Figure 1 depicts two graphics of fuzzy sets: a) with

Type-1 fuzzy logic, and b) with Type-2 fuzzy logic.

Figure 1: Difference between type-1 and type-2 fuzzy logic

membership functions.

The FLS process is divided in four parts: fuzziﬁer,

rule base, fuzzy inference engine, and output proces-

sor. In type-2 a type reducer is needed in the output

processor to derive a type-1 set from the type-2 set

(Cai et al., 2007). Figure 2 shows a diagram with a

more detailed example of a FIS structure.

Figure 2: Fuzzy Inference System diagram.

3.1 Mamdani and Takagi-Sugeno FLSs

Mamdani and Takagi-Sugeno FLSs are popular and

are formed by IF-THEN rules with the same an-

tecedent structures. The difference between them is

in the consequent structures. The consequent of a

Hybrid-IntelligentMobileIndoorLocationUsingWi-FiSignals-LocationMethodUsingDataMiningAlgorithmsand

Type-2FuzzyLogicSystems

611

Mamdani rule is a fuzzy set while in TSK is a func-

tion, so TSK uses fewer fuzzy rules to represent a

real system than Mamdani (Cai et al., 2007). TSK

FLS was proposed in an effort to develop a system-

atic approach to generating fuzzy rules from a given

input-output data set. This model consists of rules

with fuzzy antecedents and mathematical function in

the consequent part. The antecedents divide the input

space into a set of fuzzy regions, while consequences

describe behaviour of system in those regions (Ren

et al., 2006).

4 JT2FIS FRAMEWORK

JT2FIS is a class library developed for Java. The

main purpose is to deploy a library for building in-

terval Type-2 fuzzy inference systems with an object-

oriented programming language. A fuzzy inference

system (FIS) is based on logical rules that can work

with numeric values or fuzzy inputs; these rules and

individual results are evaluated together to form a

fuzzy output, then, a numerical value must be passed

through a process of defuzziﬁcation if necessary. Be-

cause it is developed in native Java it is possible to in-

tegrate its methods with Android capabilities and the

smartphone resources.

4.1 JT2FIS Clustering

JT2FISClustering is a class library developed for

Java. The main purpose is to deploy a library to

build interval Type-2 fuzzy inference systems with

an object-oriented programming language from data

mining process. A clustering method is a data min-

ing classic technique used to discover fuzzy sets and

rules to congure the FIS from real data. The library

implements a Fuzzy C-Means or Subtractive Cluster-

ing algorithm for data mining.

5 CASE OF STUDY

There is an interest on detecting the location of peo-

ple in indoor spaces, such as shopping malls and mu-

seums, in order to generate statistics about the inter-

ests of the visitors in order to guide the new ideas and

improvements of the place to cover those estimated

interests. Another advantage of knowing the location

of a person is to show them relevant or interesting in-

formation related to that place and the proﬁle of the

user. Therefore, this case of study proposes the loca-

tion of children inside an interactive museum with an

android mobile device using wireless signals from 3

different Access Points (APs) in the area. The case

of study was developed in an interactive museum ori-

ented for kids. This museum has three active ﬂoors

with different rooms on each one. The collected data

was taken in one room on the ﬁrst ﬂoor with approxi-

mately twenty interactive modules about science. For

experimental purposes the test was divided in two

phases, the ﬁrst with three not so close areas (darker

grey). The second collection of data was in four zones

(lighter grey) with few meters of distance (1 meter as

the minimum distance) as the representation shown

in Figure 3. Moreover, 3 APs are strategically posi-

tioned for getting enough coverage of the zones to do

the triangulation and estimation of the location of the

kids.

Figure 3: Case of study: zones and Access Points stablished

in the museum room.

The objective is to have enough RSSIs values from

the selected APs that describe each desired zone; then

a clustering method is applied in order to generate the

Mamdani or TKS Type-2 FISs to determine the lo-

cation. An important detail is that the location out-

side and between the selected zones is possible be-

cause when the membership of two outputs is approx-

imately equal and greater than the other output zones,

means the child or device is between the two of them.

For example, the case of the output of approximately

50% in zone 1, the output of zone 2 of about 50%, and

a low membership of zone 3, results in a position in

the showroom, meanwhile a 50% in zone 2 and zone

3, could be a zone between them that is the corridor

or the hall. And so, a collection and training of data

in every area from the place of study is not needed.

5.1 Collecting Location Data

The proposed mobile application uses the JT2FIS

class library to implement a fuzzy inference system

(FIS) and Fuzzy C-Means or Subtractive clustering

methods to do the estimation of the location process.

Wireless signals change because of different factors,

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

612

then the environment conditions when the values for

generating a speciﬁc FIS were taken, should be as

similar as possible to the time when the location was

done, or should do a new FIS for best results. The ap-

plication detects the wireless signals from at least the

three selected APs in the area to have enough sam-

ples of each different zone from each AP. There are

required at least three different APs to be able to do a

correct triangulation on the space. No stablished con-

nection in any moment is required with the APs, it is

only needed a scan of the RSSIs of the selected APs.

The collected data was taken in different zones using

a function to do a Wi-Fi scan with an Android device

searching RSSIs of three selected APs. The collec-

tion of every set of data from each zone was based on

a deﬁned limited time. Some factors, such as the total

area of the zone, determine how many data you will

need to have a good coverage of valid points for that

zone. More quantity and diversity of data can help

to have less uncertainty to generate and have a better

evaluation of the FIS. Table 1 shows samples of data.

In this example, three different inputs obtained for the

collected data correspond to each of the three AP sig-

nals. At the same time, three different outputs were

generated (in-line during collection) with a value of

one in the corresponding zone the input data was col-

lected. These were saved as .cvs ﬁles.

Table 1: Sample of collected data for each zone and the

generated outputs.

Zone Input1 Input2 Input3 Output1 Output2 Output3

1 -78 -74 -54 1 0 0

1 -80 -72 -45 1 0 0

1 -72 -66 -58 1 0 0

2 -76 -60 -58 0 1 0

2 -74 -59 -59 0 1 0

2 -65 -56 -56 0 1 0

3 -65 -66 -64 0 0 1

3 -66 -67 -60 0 0 1

3 -68 -65 -67 0 0 1

... ... ... ... ... ... ...

5.2 Mining the Collected Data

Once the inputs and outputs in location are obtained,

the next step is to use the JT2FIS class library speciﬁc

methods inside the Android application to do the data

mining in-line. The ﬁrst thing to do is the clustering,

which consists in detecting the different sets or groups

of data in all the collected data. As explained before,

each cluster found is a fuzzy rule that describes the

location system. Generally more fuzzy rules give us

a more descriptive behavior; also an increase on the

number of inputs and outputs increases the number of

rules and the complexity of the system too. Specif-

ically in location where zones are closer, a more de-

tailed description of the data is needed, thus increas-

ing the number of clusters may help. This can be con-

trolled by the FCM parameter number of clusters or

with the inﬂuence radio of each center of clusters in

Subtractive clustering method.

5.3 Results of Three Not so Close Zones

The ﬁrst tests were executed in zones with more dis-

tance between them (2 meters approximately) as the

darker zones seen in Figure 3. Table 2 shows a com-

parison of results using Fuzzy C-Means clustering

method and Mamdani FLS. Increasing the number of

clusters will increase the regression coefﬁcient that

indicates the similarity grade of the evaluation data

with the target real data. Also increasing the number

of clusters increases the time of generation and evalu-

ation of the FLS. Taking an acceptable regression co-

efﬁcient of 0.85, the less rules cost was with a cluster

number of 4 with an acceptable coefﬁcient of 0.8677.

This evaluation was generated by 1680 data per input

in total of the different 3 zones.

Table 2: Comparison table Mamdani 3 inputs 3 outputs.

Clusters Gen. time (secs) Eval. time (secs) Regression coeff.

3 5.09 0.03-0.1 0.7041

4 3.94 0.07-0.121 0.8677

7 20.41 0.07-0.12 0.8881

10 37.27 0.094-0.18 0.8913

15 56.86 0.14-0.21 0.8918

30 97.45 0.26-0.34 0.9005

70 161.66 0.61-0.97 0.9061

On the other hand, Table 3 shows Takagi-Sugeno

data with the same values used with FCM with Mam-

dani. In this case the modiﬁed parameter was the in-

ﬂuence radius or granularity grade from 0.9 to 0.1. A

low radius tends to generate more clusters; the num-

ber of clusters is the number of rules obtained. A

low radius implies clusters with closer data. High ra-

dio implies clusters with dispersed data or low den-

sity, so generally will obtain a fewer number of clus-

ters. Also, the approximation to the clusters process is

slower in subtractive method than FCM, but even with

the same number of found clusters a better regression

coefﬁcient is obtained with subtractive and TSK than

FCM and Mamdani. Meanwhile, the elapsing time of

evaluation decreases on TSK because the operations

of the consequent parts are mathematical equations

rather than Mamdanis fuzzy sets operations.

5.4 Results of Four Closer Zones

Another test was developed in four zones with less

Hybrid-IntelligentMobileIndoorLocationUsingWi-FiSignals-LocationMethodUsingDataMiningAlgorithmsand

Type-2FuzzyLogicSystems

613

Table 3: Comparison table Takagi-Sugeno 3 inputs 3 out-

puts.

Radius Clusters Gen. time (secs) Eval. time (milli secs) Regression coeff.

0.9 4 217.96 3.1 - 3.8 0.7361

0.7 4 217.91 3.1 - 5.2 0.9143

0.5 4 220.90 3.3 - 5.5 0.9245

0.4 4 224.36 4.2 - 5.1 0.9307

0.3 4 235.08 3.1 - 4.0 0.9345

0.2 8 222.05 7.0 - 10.0 0.9528

0.1 19 242.30 12.4 - 17.8 0.9741

distance between them, as shown in the lighter grey

zones in Figure 3. This set of data is about 1344 data

for each input and for each zone or output. Is difﬁcult

to identify the grade of membership of the points be-

cause of the low distance between each zone (about

1 meter approximately). As shown in Table 4, even

with 150 clusters in FCM, the regression coefﬁcient

is very low and the evaluation time cost is more than

2 seconds in that case.

Table 4: Comparison table Mamdani 3 inputs 4 outputs.

Clusters Gen. time (secs) Eval. time (secs) Regression coeff.

4 10.6563 0.0035 - 0.0041 0.3296

5 15.7962 0.069 - 0.145 0.3680

10 30.7855 0.15 - 0.23 0.3969

30 70.7959 0.39 - 0.56 0.4573

... ... ... ...

100 155.2 1.42 - 1.5 0.4264

150 165.8371 2.0 - 2.32 0.4776

Table 5 shows information of the generation and

evaluation of the set of data using subtractive clus-

tering method and TSK FLS. To achieve the desired

regression coefﬁcient, a lower inﬂuence radius was

needed (0.1), however was not able to be successfully

executed in the test device. The regression coefﬁcient

with a 0.1 radius was obtained with Matlab tools, and

the number of rules obtained was 153. Again the eval-

uation time is slower in Mamdani rather than TSK

FLS, but subtractive is slower than FCM.

Table 5: Comparison table Takagi-Sugeno 3 inputs 4 out-

puts.

Radius Clusters Gen. time (secs) Eval. time (milli secs) Regression coeff.

0.9 4 154.4764 0.0560 - 0.17 0.5237

0.7 4 155.6307 0.0390 - 0.022 0.5219

0.5 4 158.0025 0.0088 - 0.0037 0.5173

0.4 6 154.6215 0.0052 - 0.007 0.5343

0.3 12 161.7910 0.01100 - 0.016 0.5874

0.2 31 198.2 0.021 - 0.048 0.7475

0.1 - - - 0.9875

5.5 Information of Devices

The device to sense the RSSIs from each AP and

where the tests were executed is a Samsung Galaxy

Tab 4 7.0, 1.4Hz Quad Core Processor, 4KmAh

Battery. The three routers or APs used were a

Belkin Wireless G Router 2.4 GHz-802.11g model

no. F5D7230-40, an AirPort Extreme by Apple model

no. A1034 and a D-Link Wireless Router 2.4 GHz-

802.11g model DI-524.

6 CONCLUSIONS

This paper introduced a location on indoor areas

method using Wi-Fi signal networks, Type-2 fuzzy

inference system approach with a clustering method,

and how this can help to make applications that help

users in real time depending on the context. On

this approach, data was collected by one mobile de-

vice that sensed Wi-Fi signals from real indoor ar-

eas. Then, we built a Java Type-2 fuzzy inference

system using C-Means and Subtractive clustering al-

gorithm. The obtained fuzzy system can be Mam-

dani or TSK type which is later used to evaluate fu-

ture signals on mobile devices in order to decide the

current localization area. FCM is quicker than sub-

tractive but it is recommended only when having an

idea of the patterns in the collected data, otherwise is

complicated to estimate an acceptable number of clus-

ters to have an acceptable FLS structure. Other types

of pre-processing techniques can be implemented to

solve this problem. Subtractive is slower to identify

the centers but gives better results once possible opti-

mal centers are found. An expert is not necessary to

estimate the number of clusters; in less iterations it is

possible to ﬁnd optimal centers. For evaluation, TSK

is better than Mamdani because it is desired to esti-

mate the location of the device in real time, and eval-

uation in Mamdani implies more processing becom-

ing slower. As future work, we are planning to use

a neuro-fuzzy system in order to improve precision

on how to produce the fuzzy inference system from

data. Also a found limitation of this proposed method

is the distance between the zones to locate. It is not

implementable in mobile devices when the distance

between zones is approximately about less than 1.5

meters. There are details to test using Wi-Fi devices

about variety of the signals depending on the number

of persons in the places, or position of objects, and

temperature, also the impact of adding one more in-

put (AP) to the system. This information is unknown

in this work; actually there are considerations to im-

prove about lost signals of APs during location, so a

strategy to solve this problem is still in construction.

Finally for an evaluation of the method, a regression

between target and outputs values gives a better idea

of the behavior of the FLSs depending on the param-

eter of the clustering methods, but a ﬁnal and correct

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

614

evaluation of the system with a confusion matrix will

achieve this goal. Also to generate the FISs and eval-

uate them in this work the same values were used, so

it is needed to implement a thorough method that uses

a percentage of random data used in the generation of

the FISs and a percentage of real tested data. These

two samples of data will be a better set of data to test

during evaluation with a confusion matrix.

ACKNOWLEDGEMENTS

This work was supported in part by the Internal Fund

for Research Projects (Grant No. 300.6.C.135.17) of

the Autonomous University of Baja California, Mx-

ico.

REFERENCES

Bezdek, J. C., Ehrlich, R., and Full, W. (1984). Fcm: The

fuzzy c-means clustering algorithm. Computers and

Geosciences, 10(2-3):191–203.

Bozkir, S. and Sezer, E. (2013). Fuat - a fuzzy clustering

analysis tool. In Expert Systems with Applications,

FUZZYSS11: 2nd International Fuzzy Systems Sym-

posium, vol. 40.

Cai, A., Quek, C., and L., M. D. (2007). Type-2 ga-tsk

fuzzy neural network. In Evolutionary Computation.

CEC 2007 IEEE Congress.

Chai, W., Zhou, J., Chen, C., Nies, H., and Loeld, O.

(2011). Continuous indoor localization and naviga-

tion based on low-cost ins/wi-ﬁ integration. In Inter-

national Conference of Positioning and Indoor Navi-

gation.

Chiu, S. (1994). Fuzzy model identication based on cluster

estimation. In Journal of Intelligent and Fuzzy Sys-

tems, vol. 2.

Crows, T. (1999). Introduction to Data Mining and Knowl-

edge Discovery. Two Crows Corporation.

Dey, A. (2001). Understanding and using context. Personal

and Ubiquitous Computing, 5(1).

Han, J. and Kamber, M. (1998). DataMining: Concepts and

Techiques. Morgan Kaufmann, 2nd edition.

Hwang, S. and Donghui, Y. (2012). Gps localization im-

provement of smart- phones using built-in sensors.

nternational Journal of Smart Home, 6.

Jang, J., Sun, C., and Mizutani, E. (1997). Neuro-Fuzzy

and Soft Computing: A Computational Approach to

Learning and Machine Intelligence. MATLAB Cur-

riculum Series. Prentice Hall.

Navarro, E., Peuker, B., and Quan, M. (2011). Wi-Fi Local-

ization Using RSSI Fingerprinting. PhD thesis, Com-

puter Engineering Department.

Ren, Q., Baron, L., and Balanzisnki, M. (2006). Type-2

takagi-sugeno-kang fuzzy logic modeling using sub-

tractive clustering. In Fuzzy Information Processing

Society. NAFIPS 2006. Annual meeting of the North

American.

Sanchez, L., noz, L. M., Galache, J., Sotres, P., Santana, J.,

and Gutierrez, V. (2014). Smartsantander: Iot exper-

imentation over a smart city testbed. Computer Net-

works.

Vaidehi, V., Monica, S., Saafer, S. S., Deepika, M., and

Saangeetha, S. (2008). A prediction system based on

fuzzy logic. In Proceedings of the World Comgress on

Engeineering and Computer Science, WCECS 2008.

Yin, X. F., Khoo, L. P., and Chong, Y. T. (2013). A fuzzy c-

means based hybrid evolutionary approach to the clus-

tering of supply chain. Computers and Industrial En-

gineering, 66(4):768–780.

Ying, H., Ding, Y., Li, S., and Shao, S. (1998). Typi-

cal takagi-sugeno and mamdani fuzzy systems as uni-

versal approximators: Necessary conditions and com-

parison. In Fuzzy Systems Proceedings, 1998. IEEE

World Congress on Computational Intelligence. The

1998 IEEE International Conference Vol 1.

Zadeh, L. (1965). Fuzzy sets. Information and Control,

8(3):338–353.

Hybrid-IntelligentMobileIndoorLocationUsingWi-FiSignals-LocationMethodUsingDataMiningAlgorithmsand

Type-2FuzzyLogicSystems

615