CANCER DETECTION USING PRINCIPAL COMPONENT ANALYSIS AND LONG-SHORT TERM MEMORY

• Main Author: T Christopher Sirait, Daniel
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78057
• Institution: Institut Teknologi Bandung

Cancer is one of the most dangerous diseases worldwide. Abnormal cells go out of control and can invade other tissue cells wherein harmful cancer cells can spread to other parts of the body through the blood. According to WHO (World Health Organization), the biggest cause of death globally that takes 10 million lives due to cancer. The mortality rate will increase, and it is going to be fatal every year without early diagnosis. One way to detect it is to use microarray technology that monitors a very large number of expression data (genes) simultaneously. The datas used in this research are colon, ovarian and lung cancer. However, the main obstacle in a microarray data is the size of the dimensions which affects the accuracy result and time needed to process for the worse. Therefore, a plan is required to reduce such huge dimension and process it with a classification technique afterwards, so that the microarray data classification scheme can obtain good results and accuracy . In this study, CRISP-DM methodology is used to create an effective predictive model and handling out analytical data problem. Principal Component Analysis (PCA) functions as a feature extraction technique to reduce large dimensions in microarray data and applies the Long Short-Term Memory (LSTM) deep learning technique for the classification process. By using LSTM, it is proven that the accuracy value obtained is much greater and the processing time required is faster than LSTM with the help of PCA which brings down the accuracy result. The results of the classification with the best model show that LSTM can achieve the accuracy and F1 of 100% for lung cancer with time of 4164 seconds. Meanwhile, the best LSTM+PCA model obtained an accuracy and F1 of 100% for lung cancer in 4.6s.