Bayesian Inference for Crop Classification Models

This study introduces a random walk based greedy algorithm with stochastic updates for interpreting black-box machine learning model. A Bayesian hierarchical probabilistic surrogate model was formulated to quantify the uncertainty of the latent parameters of the machine learning model using a hyper-prior. Bayesian credible intervals were obtained to determine the significance of each parameter. Furthermore, the proposed algorithm is compared with the standard logistic regression surrogate model. The algorithm simulation performance was tested using Crop Recommendation based on Soil Properties and Weather Prediction Dataset.