Özet:

Abstract Cancer has emerged as a prominent issue over the last decade, necessitating timely identification for effective treatment. Consequently, the development of an automated diagnostic system for precise cancer detection has considerable significance. The analysis of microarray data is a significant challenge due to the complexity of the data, limited sample size, uneven distribution of classes, the presence of noisy data structure, and more variability in feature values. The existing Machine Learning (ML) led to lesser classification accuracy and its training is proven to be quite expensive with over-fitting problem. In contrast to the drawbacks of conventional machine learning (ML) methods like support vector machines, decision trees, logistic regression, etc., deep learning (DL) methods provide a variety of benefits including large data compatibility, automated feature engineering, and ease of use. The objective of this study is to propose the development of Mote Carlo Relief-F feature filtering (MCRelief-F) as a feature estimator that can effectively provide quality assessments of features while dealing with intricate situations characterized by significant interdependencies among features. When it comes to feature selection, SOT (scyphozoan optimization technique) is more successful than traditional approaches because it has better convergence ability, search stability, and optimum-seeking ability. The Hybrid Extensive Kernel Convolutional Neural Network (HEKCNN-LSTM-TL) uses an extensive convolution kernel for local convolution and long-term memory (LSTM) with transfer learning to improve classification accuracy while shortening training times. The suggested technique is tested on three common microarray cancer datasets, including brain, breast, and leukemia, and the feature values are scaled using artificial SOT. Classification accuracy, precision, f-measure, specificity, sensitivity, and MCC are used to assess how well the given strategy performs. A comparison of the suggested approach's performance with state-of-the-art approaches is done, and the results show that it performs better than many of the current methods, notably on the leukemia dataset.

Anahtar Kelimeler: