Hybrid Deep Learning Model with Optimization Algorithm for Precise Skin Disease Prediction and Classification

Abstract

In recent years, developing a cost effective and high efficiency screening mechanism is critical. To resolve these issues, an innovative methodology was presented in this article for identifying skin diseases. Each collected image was initially preprocessed and cropped to a size of pixels. With six square patches, these images are separated into pixels. Subsequently, image augmentation steps including flipping, rotation and image improvement are employed for minimizing the parameter requirements in further processes. The kernel weighted fuzzy local information c-means clustering (K-FCM) methodology was deployed for locating and fragmenting the cancer-affected parts. Consequently, both color and texture attributes are captured. Further, the Deep Long and Short Term Memory (DLTM) based Tunicate Swarm algorithm (TSA) was designed for identifying and categorizing skin diseases as normal or abnormal. This study was modeled in MATLAB and used the experimental image database acquired from Herlev University Hospital in Denmark for validation. As per findings of the comparison analysis, the suggested DLSTM-TSA outperforms the competition in terms ofF-score, sensitivity, accuracy andprecision.