**3.5. Cluster formation**

Cluster formation is really important to avoid cluster reconfigurations. Some of the cluster formation techniques are based on position based clustering. In these types of protocols, each road is divided into cells and in each cell some anchor points are defined. The cluster structure is determined by the geographic position of the vehicle. Another type of position based algorithm is based on hierarchical and geographical data collection and dissemination mechanism. The cluster formation is based on the position of the vehicles at a particular segment instead of the individual positions. However, this type of protocols incurs more overheads for V2V and V2I communication. In some other approaches, each vehicle entering into the network collects the neighbor vehicles information, assuming precedence to each vehicle and polls each vehicle individually (according to precedence) to check whether it is CH or not and then joins the cluster. Also every vehicle in the network collects 2-hop neighbor's information along with 1-hop neighbor's information from the CH through periodic polling. These two information collection leads to more overhead in V2V communication.

Some clustering algorithms estimate the future mobility of nodes predicting the probability that the current neighborhood of a mobile node will remain the same. The drawback of the prediction method is the lack of accuracy in some cases. In some of clustering algorithms, the clusters are formed based on mobility metric and the signal power detected at the receiving vehicles on the same directed pathway. Through such method this type of protocol helps in forming stable clusters. However, it does not consider the losses prevalent in the wireless channel. In practical scenario effects of multi path fading are bound to affect the cluster formation method and thus the stability. These effects of multi path fading are taken into account in the density based clustering algorithm. The cluster formation is based on the weight metric which takes into consideration the link quality and the traffic conditions. It can be seen that the stability is improved compared to other approaches.

In some clustering approaches considers the behavior of the vehicles, using the speed and direction parameters.

The cluster formation is based on direction of vehicle at the approaching intersection. In other approach, cluster is formed based on distance and direction of vehicle it takes after crossing the junction. Some of the research enforces a weight cluster mechanism with a backup manager. These algorithms operate in similar way. Algorithms consider the position, direction, speed and range of the nodes to perform the algorithm. On the other hand, some takes into consideration the number of neighbors based on the dynamic transmission range, the direction of vehicles, the entropy, and the distrust value parameters. They works with an adaptive allocation of transmission range (AATR) technique, where hello messages and density of traffic around vehicles are used to adaptively adjust the transmission range among them. The destination of the vehicles is used as a parameter to arrange clusters.

In some approaches, the cluster formation interval is constant, which implies a synchronous creation of clusters. This does not allow for effective cluster reorganization. The directional based clustering algorithm are based on the following mobility metrics (a) moving direction (b) leadership duration (c) projected distance variation of all the neighboring vehicles over time. In practical scenario effects of multi path fading are bound to affect the cluster formation method and thus the stability. Some approaches take into account the destination of the vehicles to arrange the clusters and implements an efficient message mechanism to respond in real time and avoid global re-clustering. There might be a problem with knowing the final destination a priori as drivers usually do not use navigation system for known routes. Some algorithms are proposed for calculating the density of vehicles in a particular region around the junction. Moreover, other algorithms groups vehicles into clusters based on the competitive learning Hebb neural network. A suitable solution to prolong the cluster lifetime, stability, fairness, avoid congestion and overhead considering the vehicular behavior is essential.
