| 摘要 |
<p>The present invention relates to a method and apparatus for predicting the value at a future time of a variable associated with an input data sequence. Such a data sequence may be a sequence of data representing the financial transactions and personal data of a customer of a business such as a bank or a retail establishment and the variable may be the profitability of the customer to the business. <??>The invention includes a database for registering a plurality of reference data sequences, for example the data sequences of an established set of bank customers. Values of the input data sequence are matched to values of the reference data sequences according to a dynamic time warping algorithm to determine the distances between the input sequence and the reference sequences. A regression function is applied to derive a weighting for each reference sequence according to its distance from the input sequence. A partial prediction of the future value associated with the input sequence is made from each reference sequence. The future value is then predicted as the sum of the partial predictions weighted by the weightings.</p> |
