Abnormal Financial Transaction Detection via AI Technology

Financial supervision plays an important role in the construction of anti-corruption and honesty, but financial data has the characteristics of non-stationary, non-linearity, and low signal-to-noise ratio, and there is no special training set that is used to identify abnormal financial data. This paper generates time series of financial transaction data with a weekly time span, and selects the total transaction amount, transaction dispersion coefficient, and the number of transfers as the characteristics of financial account data. The features are then input in a weighted one-class support vector machine (WOC-SVM) model to determine whether the transaction is abnormal. The weighted one-class support vector machine (WOC-SVM) is learnt on a training set which consists of massive normal transaction due to the difficulty to collect abnormal transactions. The parameters in WOC-SVM are tuned by cross-validation. The experiments on simulation data demonstrate the effectiveness of the WOC-SVM model learnt on selected features to detect suspicious values.