Hybrid missing data imputation and novel weight convolution neural network classifier for chronic kidney disease diagnosis

CKD (chronic kidney disease) have been identified as a serious public health concern globally. Machine learning
models can successfully enable physicians to reach this aim because of their rapid and accurate identification
performance. In this paper, KNN (K Nearest Neighbor) imputations, which choose multiple full samples with the
most comparable values for replacing missing values have been utilized in this work. Additionally, conventional
MLT (Machine Learning Technique) need to produce better outcomes than DLT (Deep Learning technique). The prediction step may be sluggish, sensitive to the size of the data, and filled with irrelevant information when there is a lot of data. Tensor factorization and ANFIS (Adaptive Neuro-Fuzzy Inference System) are added for missing data imputations to address this issue. The technique for addressing feature selections is called AWDBOA (Adaptive Weight Dynamic Butterfly Optimization Algorithm), inspired by nature. At the same time, adjustable weights for feature selection from the dataset are added. A modified form of NN called NWCNN (Novel Weight Convolution Neural Network) classifier uses convolution rather than standard matrix multiplication in at least one of its layers. Convolution layers are hidden layers in NWCNN, and kernel functions improve or fine-tune the classifier’s parameters. The dataset of CKD was imperturbable from UCI (University of California, Irvine) ML
repository and had a large number of misplaced values. This work’s proposed technique is evaluated in terms of
precision, recall, F1-score, sensitivitiy, specificity, and accuracy with the values of 99.17%, 98.71%, 98.94%,
98.71%, 99.10%, and 99.04% were obtained which is higher than the other models.