Cite: S. ‘Uyun, “Feature Selection Based on Multi -Filters for Classification of Mammogram Images to Look for Signs of Breast Cancer”, KINETIK, vol. 7, no. 3, Aug. 2022. https://doi.org/10.22219/kinetik.v7i3.1437 Kinetik: Game Technology, Information System, Computer Network, Computing, Electronics, and Control Journal homepage: http://kinetik.umm.ac.id ISSN: 2503-2267 Vol. 4, No. 3, August 2019, Pp. 277-288 Kinetik: Game Technology, Information System, Computer Network, Computing, Electronics, and Control Journal homepage: http://kinetik.umm.ac.id ISSN: 2503-2267 Vol. 7, No. 3, August, Pp. 211-218 211 Feature selection based on multi-filters for classification of mammogram images to look for signs of breast cancer Shofwatul Uyun Universitas Islam Negeri Sunan Kalijaga, Indonesia Article Info Abstract Keywords: Multi-Filter, Feature Selection, Classification, Mammogram Article history: Received: April 15, 2022 Accepted: June 14, 2022 Published: August 31, 2022 Cite: S. ‘Uyun, “Feature Selection Based on Multi- Filters for Classification of Mammogram Images to Look for Signs of Breast Cancer”, KINETIK, vol. 7, no. 3, Aug. 2022. https://doi.org/10.22219/kinetik.v7i3.1437 *Corresponding author. Shofwatul Uyun E-mail address: shofwatul.uyun@uin-suka.ac.id The accuracy of classification results on mammogram images has a significant role in breast cancer diagnosis. Therefore, many stages consider finding the model has a high level of accuracy and minimizing the computing load, one of which is the accuracy in using the best feature. This needs to be prioritized considering that mammogram image has many features resulting from the mammogram extraction process. Our research has four stages: feature extraction, feature selection-multi filters, classification, and performance evaluation. Thus, in this research, we propose algorithms that can select the features by utilizing multiple filters simultaneously on the filter model for feature selection of mammogram images based on multi-filters/FSbMF. There are six feature selection algorithms with a filter approach (information gain, rule, relief, correlation, gini index, and chi-square) used in this research. Based on the testing result using 10-fold cross-validation, the features resulting from the FSbMF algorithm have the best performance based on the accuracy, recall, and precision from 72,63%, 70,38%, 75,01% to be 100%. Furthermore, the number of resulting features is the minimum because it results from intersection operation from the feature subsets resulting from the multi-filter. 1. Introduction Mammogram image is a medical imaging technology widely used to detect and diagnose breast cancer. One of the advantages of mammogram images compared with other medical imaging technology is USG (ultrasound). In addition, the use of mammography technology does not depend on the operator's skill so that the resulting image is objective. However, analyzing an abnormality on a mammogram image is a challenging task because a mammogram image has deficient quality. Therefore, this research is trying to develop a method of Computer-Aided Design (CAD) to produce a better level of accuracy. Nevertheless, mammogram image classification can be used for screening or diagnosis [1], [2]. The feature extraction results on mammogram images produce many attributes or features. Nowadays, the data sample number or dimension development of data numbers is increasing rapidly on some learning machine apps such as pattern recognition, computer vision, or biomedical. This certainly will become a great challenge for some learning machines. The use of many irrelevant or less relevant features does not only make the learning process slower. It may result in lowering the performance of some learning tasks but will also complicate the model interoperability. Therefore, the feature selection process is an effective way to solve the problem by deleting data on irrelevant and redundant features. If the feature selection process is implemented, there are three advantages: faster computing time, increased accuracy level, and an easier way of analyzing and studying the learning method and data [3]. Feature selection finds relevant feature subsets from a set of source (original data) feature subsets. Feature selection plays the role of a data compressor on a small scale by deleting irrelevant and redundant features. Furthermore, it can also be used in preprocessing stages of a learning algorithm that is capable of producing good qualities so that it may increase the learning accuracy (the feature selection process will delete the insignificant features that may cause misleading due to overfitting; thus, the accuracy value will increase); reduce the learning time; minimize the overfitting (feature selection process may delete the redundant data and noise that may cause the overfitting on the following procedure) and simplify the learning results (the more precise the dimension of dataset is, the faster and more efficient the algorithm learning will be able to run). The feature selection process includes the combination of the search process, estimation of feature effects on data label determination, and evaluation using a machine learning algorithm. Feature selection involves many processes; therefore, it requires a heuristic search procedure to optimize the feature selection process combined with evaluator functioning to estimate the feature effect level [4]. There is a similarity between extraction and feature selection in reducing dimensions. However, both have different characters in their process. The studies on feature selection [3] have often discussed some fields such as image recognition [5], image retrieval [6], data mining [7], image mining [8], intrusion detection [9], malware classification [10], speaker identification [11], and bioinformatics [12]. Based on the use of data training, feature selection can be