By making predictions from the model created with machine learning, specifically supervised learning, this essay focuses on the auxiliary diagnosis of osteoporosis. The chosen dataset includes highly relevant indexes that determine the diagnosis of osteoporosis. After doing the feature selection and one-hot encoding to preprocess the dataset, we then split the dataset into validation and training sets to avoid overfitting. Then, to train the model for prediction, we used the neuron network, chose ReLU and Sigmoid functions as activation functions, and changed the number of neurons and epochs to find the best-fit model. Through further measurement, evaluation, and application, including plotting the binary cross entropy error diagram and analyzing the classification report, the model is examined as highly accurate and desirable. With an appreciable degree of accuracy, this model for auxiliary diagnosis is able to prevent disease development or deterioration by detecting early symptoms and vulnerable populations, reduce the probability for doctors to misdiagnose, and improve the efficiency and quality of the osteoporosis diagnosis, therefore enhancing patient experiences.
Read full abstract