Evolutionary fuzzy learning for Chinese medicine liver syndrome differentiation

As a basic principle in traditional Chinese medicine (TCM), syndrome differentiation is a comprehensive analysis of clinical information to provide evidence for treatment, which is a task that heavily relies on the subjective experience of TCM doctors and is difficult to express deterministically and quantitatively. To effectively overcome the uncertainty and improve the interpretability, we proposed an evolutionary fuzzy learning method for syndrome differentiation of the TCM liver system, which used neuro-fuzzy inference systems to infer the severity of six typical syndromes (liver depression, liver blood deficiency, liver yin deficiency, liver-fire, liver-cold, and damp-heat liver-gallbladder) based on a wide set of symptoms as input features. To determine the most appropriate fuzzy membership functions of the fuzzy learning machines, an evolutionary algorithm was employed to optimize the types and parameters of the fuzzy functions simultaneously. We compared our model with seven popular or state-of-the-art models on a real-world dataset of 11,250 samples (9000 for training and 2250 for test). In terms of regression performance averaged over the six syndromes, our model obtained the mean absolute error of 0.139, mean squared error of 0.055, explained variance score of 0.811, and R2 score of 0.811; in terms of the performance of multi-classification (none, mild, and severe classes of each syndrome), our model obtained the average precision value of 0.936, recall value of 0.96, and F-score of 0.946. Each metric value of our model was the best among the comparative models, demonstrating the performance of the proposed evolutionary fuzzy learning for TCM liver syndrome differentiation.