Noise hunting is a critical requirement for realizing design sensitivity of a detector, and consequently, noise origins and its features have been revealed partially. Among the noise sources to be hunted, sources of nonlinearly correlated noise, such up-conversion noise, are hard to find and can limit the sensitivity of gravitational wave searches with advanced detectors. We propose using a correlation analysis method called maximal information coefficient (MIC) to find both nonlinear and linear correlations. We apply MIC to the scattered light noise correlated between the seismic activity and the strain signal, which limited the sensitivity of the Virgo detector during the first science run. The results show that MIC can find nonlinearly correlated noise more efficiently than the Pearson correlation method. When the data is linearly correlated, the efficiency of the Pearson method and MIC is comparable. On the other hand, when the data is known to be nonlinearly correlated, MIC finds the correlation while the Pearson method fails completely.