The tuning of classification refinement using hierarchical grouping of categories is presented in this paper. The refinement can improve the accuracy of classifiers in the case of low confidence level and it uses a classifier, a threshold and a dendrogram as parameters. For the examination, the -NN and the Naive Bayes classifiers are used and the dendrogram will be generated by using linkage method and dissimilarity value of gravitational force-based approach on the topology information. The topology of the environment is described by IndoorGML (Indoor Geographic Markup Language) document. The data set for the classification is part of the Miskolc IIS (Institute of Information Science) Hybrid IPS (Indoor Positioning System) Data set recorded with the ILONA (Indoor Localization and Navigation) System. Three properties are examined of a setup, namely hitRate, confidence and abstraction, however, they are conflicting. A fitness function is introduced using these properties for the purpose of tuning. In this paper, the diferent weight tuples are examined in the given test environment. The goal of the paper is to examine the weighting possibilities of the hitRate, confidence, and abstraction level features for indoor positioning purposes.
These days people dependent on technology, our life has become unimaginable without high-tech tools and gadgets. We highly rely on navigation, which gives us turn-by-turn directions, trafic congestion information, and alternative routes to a given location. The demand arisen to use navigation in complex buildings like airports, railway stations or hospitals. However, classic Global Positioning Systems do not work in indoor spaces. As a result, Indoor Positioning Systems (IPS) are introduced.
Indoor Positioning Systems can be used to determine the position of people or
objects in buildings and closed areas. IPS has been considered as an active research
ifeld since the early 1990s, and these systems are detailed in the following surveys
[
Indoor positioning is challenging due to the unique properties of the indoor environment. Developers have to make trade-ofs between accuracy and cost when they choose a technology. Currently, indoor positioning is vital for smart environments. However, a suficiently precise, easily accessible, and sustainable industrial standard has not been created yet.
Symbolic positions can be considered as a category, thus the symbolic positioning can be converted into a classification problem. Some well-known classifier accept classes as prediction based on the confidence values. There are some cases when the confidence for each class is relatively small. Hence, the accuracy of these classifiers can vary in a moderate range.
For symbolic indoor positioning purposes, a classification refinement using
hierarchical grouping of categories had been proposed [
To boost the performance of the classification, a hierarchical grouping of class
categories was introduced [
The concept of enhanced classification requires parameters, namely the
classiifer, the threshold and the dendrogram. The classifier is a method for supervised
learning based on the training set and data set, where the target is a discrete
attribute. The threshold is a real value between 0 and 1, which determines whether
the prediction is accepted or the proposed concept is used. If the confidence value
of the predicted class is equal to or higher than the threshold, the classifier method
returns with the class. The dendrogram can be predefined by a linkage matrix or
it is produced by linkage [
The tree structure generated by the hierarchical clustering can be seen in Figure 1. The leaf nodes are the rooms denoted by the uuid, while the root node is the whole described environment.
The following process of the enhancement concept is performed.
1. The prediction is performed with the classifier. 2. If the confidence of the predicted class is equal to or higher than the threshold, the process terminates by returning the class as the result. 3. The leaf node in the tree is located using the uuid. 4. Until the confidence of the current node is not reaching the threshold or the root node is reached. (a) The parent of this node is selected for examination. (b) Its confidence is calculated as the sum of the confidence values of its descendant leaf nodes. 5. The process terminating by returning the contained zones of the lastly examined node.
The concept of enhanced classification requires parameters, namely the classifier,
the threshold and the dendrogram. In the experiment, the –NN and the Naive
Bayes classifiers are used to the available functionality to return the class
probabilities. These classifiers are instance-based classifier, well-known and easy to
parameterize. The –NNW denotes the weighted vote version of the –NN
classiifer in this paper. The threshold is noted as , and ∈ {0.6, 0.7, 0.8, 0.9, 1}.
In the experiment the dendrograms are generated by using linkage methods and
dissimilarity value of gravitational force-based approach [
The Miskolc IIS Hybrid IPS Data Set [
The topology of the building had been described using IndoorGML [
To narrow the scope of the experiment, the environment is chosen to be the second floor of the Miskolc IIS Building. Hence the used data set is also narrowed to 431 measurements. From the narrowed data set, the training and the test set are constructed by using stratified sampling with 0.9 and 0.1 ratio. The training and the test sets are fixed during the test. The environment contains 20 zones, and it can be seen in Figure 2. It can represent a general building with narrow corridors, a huge room, which is a lecture hall in this environment, and small ofice rooms. However, the Miskolc IIS Hybrid Dataset contains measurements taken in only 5 of these rooms, namely the East Corridor, West Corridor and North Corridor, the Lobby and the Lecture Hall 205.
Three properties are examined of a setup, namely hitRate, confidence and
abstraction. It is the rate of the correctly classified cases and all the cases to represent
the accuracy. Hence, the hitRate is a real number in the [
Confidence is a real value between the threshold and 1, including both value, which represents the accepted confidence of the result. The goal function is to maximize the confidence values.
To minimize the size of the resulted list, the abstraction feature is introduced.
However, to be consistent with the goal functions of the hitRate and the confidence,
the goal for the abstraction should also be a maximization. To eliminate the number
of rooms from the property, the level of abstraction is designed to be a real number
in the [
However, when the increment of the hitRate is focused on, the method can return all of the rooms as the result, producing a low abstraction level. In addition, when higher confidence values are aimed at increased threshold, the abstraction level can decrease. Therefore, the goal of the method cannot be based on only one of these properties. Tuning is required to find the balance of these properties to improve the enhancement of the classification based indoor positioning. − 1 ˆ = 1 − − 1
A fitness function is introduced using these properties for the purpose of tuning. The introduced fitness function assigns a non-negative weight to each property, where the sum of the weights is 1. The goal of the fitness function is to be maximized.
iftness = ℎ · hitRate + · confidence + · abstraction
The results are stored in a csv file for further processing, the schema can be seen in Table 1. The result contains 1688 rows, where the method, the , the linkage method and the weights define a setup.
ℎ
As it can be seen in Table 2, the 1 and the 1 classifiers are the most frequent, while setups using the Naive Bayes classifier is not presented. The 5 , the 5 and the 9 classifiers are presented mostly after the 1 . The average hitRate is 0.95, the average confidence is 1 and the average threshold is 0.89, and these values are not afected by the used linkage method. However, the average abstraction varied with diferent linkage method, which is shown in Table 3. method 1nn 1nnW 5nn 5nnW 9nn 9nnW 11nn 11nnW 13nn 13nnW Total Result 1nn 1nnW 5nn 5nnW 9nn 9nnW 11nn 11nnW 13nn 13nnW Total
As Table 3 shows, the average abstraction of 1 and 1 classifiers are obviously 1, while the second best value is in the case of 5 and 5 with 0.78. In the point of view of the linkage method, the weighted linkage method resulted in 0.89 average abstraction, while the last in the order is complete linkage with 0.85. The overall average abstraction of the highest fitness valued cases is 0.87.
The distribution of the average hit among the best cases according to the threshold values can be seen in Table 4.
It can be seen from the data in Table 4 that until 0.9 threshold, classifiers could not reach the highest presented fitness value with the exception of 1 and 0,6 0,9 0,9 0,7 0,9 0,9 0,8 0,9 0,9 0,9 0,9 0,9
1,0 0,9 0,9 0,9 1,0 1,0 1,0 1,0 0,9 0,9 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1nn 1nnW 5nn 5nnW 9nn 9nnW 11nn 11nnW 13nn 13nnW
Total Result 1 . With 0.9 threshold, the 5 , the 5 and the 9 classifiers could reach 1 average hit value. With the 1 threshold, only the 1 and the 1 classifier could not reach 1 average hit value.
In the point of view of the weights, the statistic made of the cases with the best presented fitness value can be illustrated in Table 5.
From Table 5 we can see that the average weight of the hit and the abstraction are similar with 0.26 and 0.24 value, while the average weight of confidence is the double with 0.5. While both hit and abstraction could be eliminated from the iftness value calculation in some cases, the weight of the confidence is at least 0.1. In the case of 1 and 1 , the hit is completely eliminated, while in the other classifiers resulted in the best fitness value presented eliminated the abstraction property.
The fitness value of the most frequent weights is presented according to the threshold and the classifier using single linkage can be seen in Figure 3. The weight for the hit and the confidence is 0.5, while the abstraction is eliminated.
As shown in Figure 3, using 0.6 threshold, the Naive Bayes, the 1 and the 1 classifiers have the highest fitness value. When the threshold is increased by 0.1, the 3 , the 3 take the lead, and 9 and 9 classifiers are also surpass the previous fitness value. However, by further increasing the threshold, the 1 , the 1 , the 3 , the 3 and the Naive Bayes classifier could not reach the highest fitness value presented. The other classifiers have increment in their fitness value while the threshold is increased. With 0.9 threshold, the 5 , the 5 , and 9 classifiers could reach the highest fitness value, while the other classifiers could reach this fitness value using 1 as threshold. 4.1. Discussion We can make observation based on three point-of-view, namely the classifiers, the threshold and the weights.
The 1 and the 1 classifiers occurred most frequently among the cases with the best fitness value presented. Contrary to the Naive Bayes classifier, which could not reach this value with any of its setups. However, the 5 , the 5 and the 9 classifiers were the second most frequent in the narrowed result set.
With given weights, the 5 , the 5 and the 9 could reach the highest iftness value with 0.9 threshold. With lower threshold, there were cases when the 1 and the 1 classifiers could reach the highest fitness value, however the average hit rate in this cases is 0.9. Using 1 as threshold, every other classifier presented in the best fitness valued setups reached 1 as average hit.
In most of the cases, only two of the three property is considered when calculating the fitness value. The 1 and the 1 classifiers neglect the hit property, while the other classifiers neglect the abstraction property. However, the weights of other two properties are equals, thus they are equally important.
The tuning of classification refinement using hierarchical grouping of categories is presented in this paper. For the examination, the –NN and the Naive Bayes classifiers were used and the dendrogram was generated by using linkage method and dissimilarity value of gravitational force-based approach on the topology information. A linear fitness function was introduced using these properties for the purpose of tuning.
The investigation of the fitness function shows that instead of three properties, the setups with the highest fitness value neglect one of the properties. The other two properties were proved equally important in the cases. However, a tested classifier could not reach the highest fitness value with any of its setups. This research has thrown up many questions in need of further investigation.
In the future, the category hierarchy enhanced classification based indoor positioning concept is planned to be examined in two ways. First is the expansion of the test environment in three dimension, which helps to test the concept in multilfoored environment. The second way is the modification of the fitness function by using non-linear elements.