Dysarthria is a communication disorder resulting from acquired progressive neurological disorders such as Parkinson’s disease, motor neuron disease, multiple sclerosis, and Huntington’s disease. Speech is formed by the acoustic excitation of the vocal tract by an air stream resulting from the lungs and pulsed at a rate that is determined by the vibration of the speakers’ vocal folds. EGG Signal of different dysarthria patients are gathered. Wavelet packet coefficient are analysed to extract the energy and entropy measures in different sub band. Based on this, value clustering is done to group the similar wave sub bands with similar lung air flow. Here feature selection is performed by using kmeans clustering based Cuckoo search algorithm (Improved cuckoo search algorithm). In the modified algorithm, modification is performed on the grouping of the attributes using kmeans clustering. Due to clustering new combinations of feature subset are created, hence can select important features very easily in shorter time.Finally dysarthria diagnosis is performed by using (GA based layered recurrent neural network (LRNN)) GA is used to optimise the weight and bias. In this work, we have included lung capacity as one of the feature in identifying dysarthria disabled person. Splitting the breathing acoustic signal captured and then computing theaverage time duration and energy of the breathing cycleis proposed in this work to estimate the lung capacity.