Bacterial contamination is a major factor affecting food safety, and the development of more accurate, rapid, and noninvasive detection techniques is important for ensuring dietary safety. In this experiment, polyvinyl alcohol (PVA)/anthocyanin (cy) nanofibrous membranes (C-PVA cy) were prepared by electrostatic spinning technology, and a prediction model for the degree of bacterial contamination was established by combining with Artificial Neural Networks (ANN) learning technique to realize the accurate prediction of bacterial concentration by color change. Scanning electron microscopy and Fourier infrared spectroscopy were used to determine the structure and composition of C-PVA cy. Then, its color responsiveness to pH and E. coli was determined to study its detection performance. Next, the ANN technique was used to learn the color change of the membrane and establish a prediction model. The C-PVA cy had uniformly thick and thin nanofiber filaments (747 nm in diameter) into which cy was successfully introduced. The fiber membrane had obvious color differences at different pH values, showing a color change from dark red to brownish red for different concentrations of E. coli, with a detection limit of 9.8 × 101 cfu/mL. A prediction model for the color values (L, a, b values) of C-PVA cy versus bacterial concentration was successfully established by ANN, with a validation accuracy of up to 96%. The C-PVA cy nanofiber color indicator film combined with ANN achieves the precise prediction of bacterial contamination level with convenient operation and high accuracy, which provides a new idea for the rapid detection of food safety.