Predicting Schizophrenia from Clinical Data with Feature Selection and CNN Logistic Regression Classification

The diagnosis of schizophrenia during early stages allows for better treatment results because of its complex debilitating nature. The treatment and diagnosis of schizophrenia early on can benefit greatly from machine learning model applications. The proposed data processing approach incorporates SMOTE (Synthetic Minority Over-sampling Technique) for addressing class imbalance together with Normalization for standardizing feature values. A hybrid model design made up of Lasso Regression for feature selection and CNN-Logistic Regression for classification helps process both imbalanced classes and features with different scales in the dataset. Lasso Regression serves as an algorithm to select important features from datasets which enhances both the interpretability and the predictive power of the model.