Generally, anomaly detection has a great importance particularly in applied statistical signal processing. Here we provide a general framework in order to detect anomaly through the statistical modeling. In this paper, it is assumed that a signal is corrupted by noise whose variance follows an ARMA model. The assumption on the signal is further compromised to encompass the inherent nonstationarity associated with natural phenomenon, hence, the signal of interest is assumed to follow an ARIMA model and the noise to denote an anomaly, however, unknown. Anomaly is assumed to possess heteroskedastic properties, therefore, ARCH/GARCH modeling could extract the anomaly pattern given an additive model for signal of interest and anomaly.