Machine learning algorithms for disease prediction in embedded systems

Today many diseases affect people as a result of lifestyle choices and the environment. Early disease prediction is very important in terms of precautions to be taken. It may not be easy for doctors to make an accurate diagnosis based on symptoms when predicting disease based on symptoms. Due to the increase in the amount of data in the field of medicine and health in recent years, it has become a very important task to predict disease from data with the help of machine learning methods. In this study, embedded system platforms; A machine learning based prediction system based on disease symptoms is proposed using Jetson Nano, Jetson TX2, PYNQ and Raspberry Pi. Data preprocessing techniques including SMOTE (synthetic data augmentation method) and oversampling methods are used to organize the obtained data set. This dataset is classified with the help of decision tree, random forest, AdaBoost classifier (ADA), gradient boosting (GB), multilayer perceptron (MLP) and linear discriminant analysis (LDA) classifiers to accurately predict the disease. The results of these classifiers are carried out for k = 20 folds using the cross validation method. In the thesis study, linear discriminant analysis (LDA) and gradient boosting classifiers (GB) outperform other classifiers in terms of output, with an accuracy rate of 97,61%. In addition, the accuracy, training and test times of the classifiers are performed in various embedded systems and the results are given comparatively. Thanks to the high accuracy and portability of the proposed system in this thesis, its usability as a decision support system in the field of medicine and health has been presented.