The the weighted k-NN model. B. Genetic

Themajor network features, considering the cellular network as an example, aredeveloping dramatically, such as the increasing demands for high data rate serviceand densely distributed traffics. Hence, network operators are facing the greatchallenges on how to improve the network capacity and minimize the coveragehole by network configuration optimization approaches in an efficient way 2. Commonnetwork optimization strategies, which passively regulate the networkparameters centered on network’s congestion ratio, drop off cost, protectionholes and etc., are no longer applicable in the dynamic changing wirelessnetwork environment. Besides, due to the variety of services and humanactivates, the radio resource shortage and service demand are distributedunevenly in different areas and fluctuate drastically over different timeperiods.Previousnetwork optimization technologies passively modify the network configurationssituated on network’s congestion ratio, drop-off rate, protection holes and soon.

So to optimize the network configurations by obtaining the accurate networkstatus, user demand and application request distribution based on the real timedata.The proposed radio resourceoptimization architecture as shown in figure 1.  Figure 1:Radio Resource Optimization ArchitectureA.   Traffic Load Prediction ModelThe For the classification task 16,the k-NN approximation model is effective and simple, because samples withsimilar inputs yield similar outputs.

In this model, the Euclidean distance isapplied as a metric to select the nearest neighbour. And the estimation of theoutput is the weighted average value of all k nearest neighbours, which isdefined as the weighted k-NN model. To be specific, if the data set is composedof (xi, yi), where xi is an n-dimensional input value and yi is a scalar outputvalue, then the weighted k-NN model. B.   Genetic Algorithm based Weighted k-NN ModelBoth input selection and k valuedetermination are key issues to construct the weighted k-NN model. The value kis decided by diverse model structure selection techniques, such as m-foldcross validation, leave-one-out (LOO) and Bootstraps.

In this paper, the m-foldcross validation is used to select the best value k. To obtain the best inputset, the exhaustive search scheme among all possible combinations of inputs istoo time consuming and complex to be used in practice. If L is the number ofcandidate inputs, there will be input combinations. To reduce the computationalcomplexity, a mixed genetic and cross-validation algorithm is proposed toobtain the optimal input set and k value for the weighted k-NN model.C. Dynamic Resource Reconfiguration Optimization   Tomake a better use of radio resource margins temporally and spatially over thewhole network, both the dynamics of the traffic load in each cell and theresource reconfiguration are vital and indispensable, which are two maincomponents of the proposed dynamic resource reconfiguration framework as shownin Figure 2. First, the daily peak-hour traffic load is recorded.

Then,historical records with diurnal patterns are used to train the k-NN model whichcan predict the traffic load for the next day. Finally, the daily peak-hourtraffic load is predicted by using the k-NN model, and the radio resources arereconfigured daily over the whole network by using the optimization algorithmsto meet resource demands in each cell. In practice, the resourcereconfiguration deployment in cellular networks can be accomplished remotely byusing the software in 15.

Therefore, the channel utilization of the wholenetwork can be improved when the traffic load balance is achieved. The proposedweighted k-NN model and resource reconfiguration optimization algorithm canalso be applied to other cellular networks with minor modifications. And thetime granularity of resource reconfigurations can be adjusted dynamically tomeet practical demands hourly or weekly. In this paper, the resourcereconfiguration is performed daily considering about the trade-off between theefficiency of resource utilizations and the cost of frequent resourcereconfigurations.

Figure 2. Dynamic ResourceReconfiguration Framework  IV.ALGORITHM Algorithm: Inputand Structure Selection Algorithm for weighted k-NN Model Input:L: length of individual, Pc: cross-over probability,GEN:maximum generation, Pm: mutation probability,POP:population size, ER: elite ratio of each generation,S:value range of genes in the individual,N:maximum value of k in weighted k-NN modelOutput:m: the optimal number of neighbors in weighted k-NNmodel, I: the best input setInitialize:m ? 0, I ? Random Initialization1: fork = 1 to N do2: rungenetic algorithm using cross validation erroras the fitness of individual, andget the best individual I(k)   and itscorresponding fitness E(k)3: endfor4: m ?arg min {E(k)}5:decode I(m) to get the best input set I6:return m and IV.ConclusionsIn this paper we havepresented the design of a radio resource optimization using k-NN Algorithm.

Ak-Nearest Neighbours (k-NN) model is proposed to predict periodiccharacteristics of network traffics and maximizes the utilization of network asper user demand or priorities of application. By applying proposed optimizationalgorithm throughput of network by increases and the radio resources can bereconfigured actively to satisfy the dynamic pattern of traffic loads with theaid of the usage of the proposed a Radio ResourceOptimization using k-NN Algorithm.