Assuming 3 segments and 5 points to distribute among them based on their KPI obtained, KPI = (2.5... more Assuming 3 segments and 5 points to distribute among them based on their KPI obtained, KPI = (2.52, 1.51, 0.97). If we simply do the assignment by rounding this KPI from the highest to the lowest, KPIr = (3, 2, 1), the first segment would have 3 points, the second 2, and there would be no more points left to assign to the third segment. The final point assignment would be P = (3, 2, 0). Although logic indicates that in the third segment there is a lot of uncertainty (0.97), it would not touch any point. This leads to 2 segments with an over-allocation of points with respect to their KPI. The over-assignment obtained would be (3-2.52 2-1.51) = 0.97 In the heuristic of 2 passes, this effect is minimized as follows: First, the integer values of KPI = (2.52, 1.51, 0.97), KPIe = (2, 1, 0) would be assigned, making an initial point assignment P = (2, 1, 0), leaving a KPI pending allocation of KPIp = (0.52, 0.51, 0.97). There would be another 2 points to allocate distributed in values <1. To minimize over-allocation, they would be assigned by order of the pending KPI, so 1 point will go to the third segment and the other to the first Pp = (1, 0, 1) being the allocation of points per definitive segment P = (3, 1, 1). Now the over-assigned segments would be the first and the third (3-2.52 1-0.97) = 0.51, this being the way to distribute the 5 points with the smallest possible signage. In this way, the algorithm adjusts the logic that is more interesting and we will obtain more information (we will lose less when over-allocating where it is not so necessary) assigning a point to the second segment and another to the third (3, 1, 1), than 2 to the second (3, 2, 0).
2005 ICSC Congress on Computational Intelligence Methods and Applications
The results of feature selection methods have a great they usually obtain better results. The hyb... more The results of feature selection methods have a great they usually obtain better results. The hybrid model attempts to influence on the success of data mining processes, especially take advantage of the two models by exploiting their different when the data sets have high dimensionality. In order to find evaluation criteria in different search stages. the optimal result from feature selection methods, we should check each possible subset of features to obtain the precision nependetlof the ealua function, the fe seon classification, i.e., an exhaustive search through the search lection methods must carry out a search among the different space. However, it is an unfeasible task due to its computational candidates of features subsets. The search can be [2]: complete complexity. In this paper we propose a novel method of feature [10], sequential or random. The complete search (or selection based on bootstrapping techniques. Our approach shows exhaustive) guarantees to find the optimum result according that it is not necessary to try every subset of features, but . . . only a very small subset of combinations to achieve the same ex evaluation criti In trat it is utationaly performance as the exhaustive approach. The experiments have expensive (O(2n)), which makes that it is unapproachable been carried out using very high-dimensional datasets (thousands when the number of features (n) is large. The sequential of features) and they show that it is possible to maintain search, with cost O(n2), is not complete and might not find the precision at the same time that the complexity is reduced the optimal subsets, because it is based on previous ranking, substantially. which has been produced by other techniques([7], [9], etc.).
Uploads
Papers by Norberto Diaz