22
Internet of Things (IoT)
and communicate to each other through synapses, dendrites, and axons. The dendrites
of a neuron act as a set of inputs while, in contrast, the axon acts as the neuron’s output.
The synapse is a junction that bridges the communication between one neuron’s output,
the axon, and another neuron’s input, the dendrites.
Artificial neural networks work by mimicking these biological processes, using a
network of artificial neurons. The neurons are based on their input variables and their
assigned weights. The weight corresponds to the strength of the synapse in the network.
If the input values meet the required level to fire, the neuron will pass on an output value
to the next set of neurons. This process repeats until the final neuron is fired upon and
outputs the classification. Both biological and artificial neural networks learn by changing their weight values, which is the strength of the synapses in biological networks [17].
For supervised learning, the back propagation algorithm is used to quickly train artificial neural networks and adjust the weights to minimize the error between predicted and
desired outputs [17,18].
2.4 Semi-Supervised Learning with Genetic
Algorithms and Rough Set Theory
2.4.1 K-means Clustering
The K-means algorithm is provided with the number of clusters k to find, which also
determines the number of centroids used [19]. A centroid is the average position of all the
points within the cluster. The K-means algorithm starts with random centroids as initial
guesses, and then determines the optimal and stable centroids for the clusters through
subsequent iterations.
The k initial centroids can be selected randomly from the dataset, or created as random locations, within the input space. Each of the input data instances are then assigned
to the closest centroid. The closest centroid is determined by a distance metric, such as
Euclidian distance. After each instance has been assigned to a centroid, a new centroid for
each cluster is calculated as the average of all of the instances assigned to it.
This process is repeated and the entire dataset is reassigned to the new centroids.
The algorithm will repeat until the cluster centroids do not change with subsequent
iterations [19].
2.4.2 Adaptation of Rough Set Theory for Clustering
Rough set theory represents a set of both lower and upper approximations instead of the
traditional nonoverlapping sets. In conventional nonoverlapping sets, the boundaries of
the sets or clusters are not always clearly defined, and objects may be equidistant from the
center of multiple clusters. In contrast, rough sets are more flexible because they allow overlapping clusters and are also less descriptive (specific) than fuzzy sets. The lower approximation of a rough set is a set comprising only the elements that definitely belong to the
subset. Objects in this set are located in the positive region. The upper approximation of a
rough set comprises elements that both definitely belong, and those that possibly belong to
the subset. The lower approximation is a subset of the upper approximation. Elements that
are outside the upper approximation are located in the negative region, whereas elements
Internet of Things (IoT)
and communicate to each other through synapses, dendrites, and axons. The dendrites
of a neuron act as a set of inputs while, in contrast, the axon acts as the neuron’s output.
The synapse is a junction that bridges the communication between one neuron’s output,
the axon, and another neuron’s input, the dendrites.
Artificial neural networks work by mimicking these biological processes, using a
network of artificial neurons. The neurons are based on their input variables and their
assigned weights. The weight corresponds to the strength of the synapse in the network.
If the input values meet the required level to fire, the neuron will pass on an output value
to the next set of neurons. This process repeats until the final neuron is fired upon and
outputs the classification. Both biological and artificial neural networks learn by changing their weight values, which is the strength of the synapses in biological networks [17].
For supervised learning, the back propagation algorithm is used to quickly train artificial neural networks and adjust the weights to minimize the error between predicted and
desired outputs [17,18].
2.4 Semi-Supervised Learning with Genetic
Algorithms and Rough Set Theory
2.4.1 K-means Clustering
The K-means algorithm is provided with the number of clusters k to find, which also
determines the number of centroids used [19]. A centroid is the average position of all the
points within the cluster. The K-means algorithm starts with random centroids as initial
guesses, and then determines the optimal and stable centroids for the clusters through
subsequent iterations.
The k initial centroids can be selected randomly from the dataset, or created as random locations, within the input space. Each of the input data instances are then assigned
to the closest centroid. The closest centroid is determined by a distance metric, such as
Euclidian distance. After each instance has been assigned to a centroid, a new centroid for
each cluster is calculated as the average of all of the instances assigned to it.
This process is repeated and the entire dataset is reassigned to the new centroids.
The algorithm will repeat until the cluster centroids do not change with subsequent
iterations [19].
2.4.2 Adaptation of Rough Set Theory for Clustering
Rough set theory represents a set of both lower and upper approximations instead of the
traditional nonoverlapping sets. In conventional nonoverlapping sets, the boundaries of
the sets or clusters are not always clearly defined, and objects may be equidistant from the
center of multiple clusters. In contrast, rough sets are more flexible because they allow overlapping clusters and are also less descriptive (specific) than fuzzy sets. The lower approximation of a rough set is a set comprising only the elements that definitely belong to the
subset. Objects in this set are located in the positive region. The upper approximation of a
rough set comprises elements that both definitely belong, and those that possibly belong to
the subset. The lower approximation is a subset of the upper approximation. Elements that
are outside the upper approximation are located in the negative region, whereas elements