Abstract
In recent years, due to an increase in the incidence of different cancers, various data sources are available in this field. Consequently, many researchers have become interested in the discovery of useful knowledge from available data to assist faster decision-making by doctors and reduce the negative consequences of such diseases. Data mining includes a set of useful techniques in the discovery of knowledge from the data: detecting hidden patterns and finding unknown relations. However, these techniques face several challenges with real-world data. Particularly, dealing with inconsistencies, errors, noise, and missing values requires appropriate preprocessing and data preparation procedures. In this article, we investigate the impact of preprocessing to provide high-quality data for classification techniques. A wide range of preprocessing and data preparation methods are studied, and a set of preprocessing steps was leveraged to obtain appropriate classification results. The preprocessing is done on a real-world breast cancer dataset of the Reza Radiation Oncology Center in Mashhad with various features and a great percentage of null values, and the results are reported in this article. To evaluate the impact of the preprocessing steps on the results of classification algorithms, this case study was divided into the following 3 experiments: Breast cancer recurrence prediction without data preprocessing Breast cancer recurrence prediction by error removal Breast cancer recurrence prediction by error removal and filling null values Then, in each experiment, dimensionality reduction techniques are used to select a suitable subset of features for the problem at hand. Breast cancer recurrence prediction models are constructed using the 3 widely used classification algorithms, namely, naïve Bayes, k-nearest neighbor, and sequential minimal optimization. The evaluation of the experiments is done in terms of accuracy, sensitivity, F-measure, precision, and G-mean measures. Our results show that recurrence prediction is significantly improved after data preprocessing, especially in terms of sensitivity, F-measure, precision, and G-mean measures.