Since your dataset is imbalanced, the tree is biased toward the majority class. Even though you tried oversampling, decision trees can still struggle with minority classes if the features don’t provide strong separability. Here are few suggestions that you should try:
1. Instead of uniform priors ('prior', 'uniform'), try setting class-specific priors to emphasize the minority class. The fitctree function has the default option of empirical which determines class probabilities from class frequencies in the response variable. Refer the following documenattion section to read more about this argument: https://www.mathworks.com/help/stats/fitctree.html#bt6cr7t_sep_shared-Prior
2. You are using classiferror in the loss function, which only considers misclassification rate. Try using cross-entropy loss ('LossFun', 'logloss') or Gini impurity ('LossFun', 'gdi').
3. If class 0 and class 1 have overlapping feature distributions, the tree might struggle to separate them. Try adding interaction terms or transforming features to improve separability. Refer the Classification Learner for this process: https://www.mathworks.com/help/stats/feature-selection-and-feature-transformation.html
4. Decision trees alone might not be the best choice for imbalanced data. Consider Random Forests (fitensemble) or Gradient Boosting Trees (fitcensemble) for better performance.
