Methods Of Research, Analysis Of Ifn-based Ticket Marketing

With the development of information technology of Chinese railway, rich ticket data have been collected in China Railway Train System (CRTTS). Making good use of the current ticket information data to acquire valuable decision information with less human and technique resources is becoming both the urgent demand for the railway decision department and the key point for the information department's work. The research of ticket center in CRTTS is still in its initial stages of statistical data, however, conventional statistical methods are difficult to establish an effective aggregation predictive models. ID3 algorithms based on passenger tickets analytical methods can establish better prediction analysis models, but models withdraw from the larger rule sets, extracting novel rules from which is more difficult for decision makers, which is the problem in CRTTS.Aimed to the problems show above, regarding the railway passenger traffic as our background, we do deeply research and experiments on IFN (Information-Fuzzy Network) algorithm to establish the corresponding data analysis models. The IFN algorithm measures the input attributes by mutual information, and selects smaller set of attributes from the perspective of overall, which can produce much more compact models than other methods of decision-tree learning while preserving nearly the same lever of classification accuracy. The model construction cost is less than other classification algorithms, which can be extracted from a few smaller association rules between input attributes and target attributes. The pre-pruning method of IFN algorithms use likelihood-ratio static to evaluate the statistical significance of the conditional mutual information between a input attribute and the target attribute. The outstanding strongpoint of the IFN algorithm's pre-pruning is that the model construction can be suspended at any time when the computing resources are limited; And the rule set obtained is not larger than the one obtained by association rule algorithm; Moreover, it does not have to suspend the net construction for the...