Traditionally diseases are assessed by looking at their Symptoms by their family general practitioners. Although this method is effective, there is a slim chance that they might misinterpret a disease and mistreat it. Mistreating could result in being harmful to individuals receiving those treatments. But, understanding the genomic structure of the individuals could help us give a better view of the diseases that are likely to happen.
We have used a Genomic sequencing dataset by Gene Ontology which has 11950 predictors of 18418 individuals. We will be predicting Heart Disease, Type II Diabetes, Alzheimer, Parkinson and Obesity. Since the target features of these diseases are imbalanced, we will be using upsampling methods ADASYN (Adaptive Synthetic) and SMOTE (Synthetic Minority Over-Sampling technique). We'll build logistic regression, random forest and neural network models for three versions of the dataset for each disease. Later the models were evaluated by looking at their Accuracy, Precision, Recall, F1 Score and ROCAUC Score using a confusion matrix.
We managed to build logistic regression, random forest and neural network model for each disease having different versions of the dataset with ROCAUC scores for the generic dataset, ADASYN dataset and SMOTE dataset between 0.49 to 0.63, 0.49 to 0.59 and 0.49 - 0.59 respectively .