Enabling Feature-Level Interpretability In Non-Linear Latent Variable Models: A Synthesis Of Statistical And Machine Learning Techniques