In this work, to overcome the problem of voluminous dataset handling, improving the relevancy of features to detect the chronic kidney disease related diabetic patients, an uncertainty handling feature selection is constructed. The dataset is initially clustered by DBSCAN, the efficiency of clustering is achieved by two parameters they are epsilon and min-points. These are assigned in a random manner in standard DBSCAN, if the value is not assigned with any knowledge, then the clustering will not produce appropriate result. Hence, this presented work adapted the Moth Flame Optimization to search for optimized parameter values using its flight path behavior It discovers the best fittest value based on the flame position and those are fine tuned. The feature's linear correlation among the class variable is identified using Pearson's linear correlation based on it's the features which doesn't belong to any cluster are eliminated and based on the correlation score they are ranked. Thus, the significant feature subset is used for chronic kidney disease detection. The uncertainty about the outliers is well handled using the moth flame metaheuristic model by finding the best fittest and worst fittest features. The simulation results also proved that the accuracy obtained by the proposed Moth Flame based DBSCAN along with Pearson Correlation (MFDBSCAN-PC) produced highest accuracy compared to other standard feature selection models in chronic kidney disease detection with reduced error rate.