الفهرس 
Title pageOnly 14 pages are availabe for public view
 |  | from | 18 |  |  |
| | | from | 18 | | |
Abstract
Cancer is the second leading cause of death worldwide. Cancer can be detected early, which lowers mortality rates and increases the likelihood of a successful therapy. The goal of this research is to provide a method for early cancer detection using machine learning. This is done by using machine learning algorithms for predicting various forms of cancer from scents in biological data. Based on electronic nose measurements, odors of biological substances.
The study comprised Egyptian men and women who were referred to the Medical Research Institute’s Hospital for diagnosis and/or treatment. All the subjects had a comprehensive medical examination, as well as demographic information and electronic nose measurements. The E-Nose employed in this investigation was a movable PEN3 (Airsense Analytics GmbH, Schwerin, Germany) with ten nonspecific metal-oxide sensors. All experiment measurements and sensor array patterns were stored in files for subsequent investigation. The data was divided into five categories depending on their health state. Controlled health participants were among the participants who showed no indications or pathologies: brain cancer, breast cancer, lung cancer, and CLL. The information came from three biological samples: blood, urine, and a biopsy.
Data is preprocessed before being split into training and testing datasets for machine learning classification algorithm training and evaluation. Support vector machine with optimized kernel and C parameter was the machine learning algorithm used in this investigation. For these data, the Optimal Training Parameter Combination (OTPC) of the SVM model was C = 200 and linear kernel.
For the blood dataset, the proposed classifier had precision, recall, F1 scores, and specificity of 98.6, 99.0, 98.6, and 98.0%, respectively. While the precision, recall, F1 scores, and specificity of the urine dataset were 99.5, 99.3, 99.3, and 99.8%, respectively. The precision, recall, F1 scores, and specificity of the biopsy dataset were 94.3, 96.0, 94.6, and 98.3%, respectively. Using the urine dataset, the proposed classifier had an accuracy of up to 99.0%.