enuzor / a-comparison-on-supervised-and-semi-supervised-machine-learning-classifiers-for-diabetes-prediction Goto Github PK
View Code? Open in Web Editor NEWThis project forked from lokesh180/a-comparison-on-supervised-and-semi-supervised-machine-learning-classifiers-for-diabetes-prediction
The thesis aims to compare the selected supervised and semi-supervised ML classifiers to predict gestational diabetes (Type-3). The chosen algorithms have been used previously by other researchers for predicting diabetes. The dataset chosen is PIMA Indians Diabetes Dataset (PIDD), consisting of female patients aged 21 and above. The PIDD dataset comprises 768 instances, of which 500 patients are non-diabetic, and the rest are diabetic (the dataset is imbalanced). Different steps such as data cleaning, feature selection, and binning are done on the dataset, which leads to two datasets, namely, non-binned and binned. Parameter tuning is performed while training the algorithms. In addition, oversampling is done on the training set to cope with the imbalanced nature of the dataset. The built models are evaluated using different performance measures. The results of the study showed that the semi-supervised classifier could perform better compare to supervised methods. The non-binned dataset seemed to be more suitable for this problem. The thesis is finely written and structured. In addition, the selected ML techniques are well explained and motivated.