Innovative Precision Medicine: An Explainable AI- Driven Biomarker-Guided Recommendation System with Multilayer FeedForward Neural Network Model

Pain, with its intricate nature and subjective character, presents significant measurement challenges. While patient self-report traditionally serves as the primary means of assessing pain, this method becomes inadequate when patients are unable to verbally express pain intensity or lack typical cognitive abilities. We propose an innovative Multilayer Perceptron (MLP)-based opioid recommendation model with Explainable AI through SHAP (SHapley Additive exPlanations). The model analyzes clinical study biomarkers and relevant metrics as part of our initiative to build a DosagePrediction Model and a PainSeverityClassification Model. These two models are seamlessly integrated, forming a unified opioid recommendation model that leverages advanced analytics to optimize pain management and enhance patient care. The MLP architecture, predicts pain intensity, medicine classification, and pain classification through a feedforward neural network with ReLU activation and a unique skip connection. This residual connection, adding the first fully connected layer's output to the second, mitigates the vanishing gradient problem during backpropagation. The training process combines mean squared error loss for regression and cross-entropy loss for classification tasks. Predicted probabilities for medicine and pain classifications guide decision-making, augmented by Explainable AI, particularly SHAP (SHapley Additive exPlanations), highlighting feature importance. Notably, our opioid dosage recommendation class achieves a remarkable 96% accuracy, while the pain severity classification class attains an accuracy of 86% and dosage prediction class attains 97%, underscoring the efficacy of our model in accurate and reliable predictions.