Network Topology in Data Mining

Network topology consists of three layers which are the input layer,  hidden layer and output layer. Before the beginning of the training of a model we must specify the number of units in the input layer, the number of hidden layers (if more than one), the number of units in the hidden layer, and the number of units in the output layer in the network topology.

Network design is a trial and error process so it may affect the accuracy of the resulting trained network. Once a network has been trained and its accuracy is not considered acceptable, it is common to repeat the training process with a different network topology or a different set of initial weights. Cross-validation techniques for accuracy estimation can be used to help decide when an acceptable network has been found. A number of automated techniques have been proposed that search for a “good” network structure.  

wireless mesh networks which is a type of network topology are efficient and low-cost solutions to expand Internet coverage to areas where infra-structured connections to IEEE 802.11access points are hard or expensive. wireless networks are expected to have high levels of usage, demanding solutions such as multichannel communications between the mesh access points to improve the throughput of wireless mesh networks. As mentioned wireless mesh networks implement a multi-channel approach which creates multiple subnetworks.

The overall network performance depends on subnetwork performances which in turn depends on topology characteristics. These topological characteristics have some impact on network performance. some metrics related to the topology characteristics identified are mentioned as follows :

• The number of nodes per subnetwork.
•  The mean hop count.
•  The neighbour node density.
•  The hidden nodes are measured by the miss ratio.
•  The number of nodes in the neighbourhood of the gateway.
•  The hidden nodes in the neighbourhood of the gateway are measured by the miss ratio.

Network Metrics:

The network metrics are categorized as performance metrices and topology metrices.

Performance Metrices: 

These metrices are characterized by using measures such as throughput, delay, packet loss, and number of retransmissions. In this we mainly focus on node throughput and delay. In network metrices, we mainly focus on the following things first maximize the average node throughput, Second maximize fairness among nodes throughput to be sure that each mesh node offers an effective connection to the clients for the infra-structured network in order to avoid node starvation while assuring fair opportunities for packet transmission. Third minimize the end-to-end delay experienced by packets transmitted between the mesh nodes and the infra-structured network. Following are network metrices :

1. Aggregate Throughput: it defines the sum of bit rate received by the destinations. The aggregate throughput T_s measured on channel s is given as  

TiRX = number of packets received by node
TiTX  = number of packets received by the gateway sent by node 
L =  L is the packet length given in bits. 

2. Fairness: The fairness J is independent of scale, applies to any number of nodes and is bounded between 0 and 1, where J = 1 indicate a totally fair network. 

3. Delay:  it defines the mean time elapsed between the creation of a packet and its reception at the final destination.

Topology Metrices:   

The number of nodes sharing a common channel is also considered since the network topology and channel assignment schemes are random.

1. Number of Nodes Difference: The difference between the number of nodes n in each subnetwork. Nodes in a subnetwork share a common radio channel and their packets are forwarded through a common gateway.

2. Neighbour Node Density : The mean number of immediate neighbours for  each node has in the network (including the gateways). 

3. Size of the first ring: The number of nodes a single hop distance from a gateway, which is also the neighbour node density that gateway.

4. Mean hop count: The mean value of the hop counts of each node in the network to reach a gateway (excluding the gateways and the nodes on the first rings which are included in the size of the first ring metric).

5. Miss ratio: Defined as  a global measure of the severity of hidden nodes in the overall network.

6. First ring Miss ratio: Defined as a measure of the severity of the hidden nodes in the neighbourhood of the gateway.