ReRAM-based Computation-in-Memory operates multiply-accumulate in parallel. ReRAM stores weights of neural network as its memory conductance. Complex fluctuation patterns in ReRAM current are converted into images of Time-lag Plots. Convolutional Neural Network (CNN) trains the artificially generated fluctuation data and classifies the measured ReRAM current data into 6 fluctuation patterns. The accuracy of Computation-in-Memory is improved by suppressing frequent ReRAM fluctuation patterns.