A machine learning classification approach for diabetes and biomedical data

People of all ages and different socioeconomic levels around the world are being diagnosed with diabetes type 2 at a higher rate than ever before. Diabetes; it is seen as an serious root cause of many important diseases such as blindness, kidney and heart diseases. For this reason, it is of great importance to establish diagnostic systems that can reliably detect diabetes based on the available medical data. In this thesis; method-based machine learning methods are proposed for classification of diabetes, heart-c, heart-h, indian-liver, vertebral-column-2c, vertebral-column-3c, parkinsons and thyroid biomedical datasets. The proposed method consists of several stages including data preprocessing, feature extraction and classification. This study focuses more on the preprocessing stage to improve the data set used in the next stages. In the mentioned preprocessing stages; firstly, the data is cleaned and the missing values are completed with the mean values. Then, data resampling is performed with the Dara method and then data normalization is applied. Finally, Principal Component Analysis (PCA) is used for feature extraction . 4 different methods for classification after preprocessing; J48, KNN, RF and DT are used. The results are obtained by using the 10-fold cross validation method to evaluate the achievements. In this way, 97.8% accuracy, recall and precision results were obtained for the PIMA diabetes dataset with the RF algorithm. Based on the results obtained the proposed method can be used to develop an effective and efficient machine learning system that can classify diabetes with high accuracy. In subsequent experimental studies on biomedical datasets, the performance of the proposed method is tested, and it is seen that 99% accuracy, sensitivity and precision results are obtained for the parkinsons dataset.