Handling outliers and over fitting for lung cancer disease prediction through neural network clustering

Abstract

With the rapid increase of cancer disease-affected people worldwide, predicting the cancer disease patients and diagnosing the cancer disease becomes crucial. The cancer disease diagnosis models have greatly confronted the data scarcity and often deal with the outliers. This paper presents the Neural network Clustering-based Outlier Handling (NCOH) methodology to improve the performance of multi-class cancer detection by modeling the outlier detection and outlier handling processes. The proposed methodology enhances the data augmentation process and the outlier handling process to enforce the cancer disease diagnosis using the oversampling and the neural network clustering methods. The NCOH approach handles the outliers with the impact of the clustering and the Z-score normalization score. The experimental results illustrate that the proposed approach significantly outperforms the existing outlier detection method by 15% of higher recall while testing the performance of the multi-class cancer classification model.