摘要 |
A wavelet based method for fraud detection is disclosed. The method includes entering time series data profiles into a feature extractor wavelet transformer (3), providing a wavelet decomposition, entering the wavelet decomposition into a processor (7), combining the processed wavelet co-efficients with the raw data profiles and assembling an extracted feature data output (21). The extracted data and data indicative of fraud history and customer labels is entered into a model generator (4) to be and classified (12)and then validated (19). Model data is then passed to an input of the fraud detector (5) together with fraud history data and the extracted feature data. Weights are allocated (22) to the extracted features and linearly combined (23). The combined output is classified together with the raw data of the fraud history to provide an output that is a cross combination (25) of the raw data and the data which is indicative of the probability of fraud. |