The invention discloses a breast cancer condition diagnosis method based on big data and machine learning. The method comprises the following steps of: preparing a substrate;, a large number of X-rayand ultrasonic images are encrypted and stored through the Hadoop distributed parallel operation platform; a filter is randomly selected from the two-dimensional image information for convolution training; on the basis, the weight on the filter is modified to enable the filter to identify a certain feature, a convolutional neural network with a high-order convolutional layer is constructed, neurons are encoded into completely relevant spatial features after an image passes through a plurality of convolutional layers, and the features can be learned by a full connection network. Through pre-feedback learning of the convolutional neural network, classification of breast cancer conditions is realized, benign or malignant breast cancer conditions are classified, and medical breast cancer falsepositive diagnosis cases are reduced.