I am currently working on a college project where the goal is to classify audio recordings as depressed or not. I have got 74 acoustic attributes assigned continuously for each recording where the recording is segmented every 10ms. So for every recording I have several thousands data points with 74 acoustic attributes. In total I have 106 recordings and output labels for each of them.
I found Bayesian approach appropriate for this project as uncertainty in the prediction is crucial. It is also as excuse to learn about this field as I am really new to it.
I can’t seem to understand how should I build the model. I know I have to use bernoulli dist to represent the output and probably a beta dist for the p value as prior. But where should I include the input datasets.
How should I approach this one?