Intelligent Data Analysis: An International Journal

Two data mining techniques were compared for their ability to improve the prediction of abnormal returns using insider stock trading data. The two were neural networks (NN) and Multivariate Adaptive Regressive Splines (MARS). In the comparison, both analyzed abnormal stock market returns from the same 343 companies over the identical 4 1 2 year period (1/93-6/97). The major findings were: 1) both NN and MARS generally identified the same industries that had the most predictive abnormal stock returns 2) both found that predictions further in the future (12 and 9 months ahead) were more accurate than predictions closer to the trading date (6 and 3 months ahead) 3) both obtained better predictive accuracy using four – rather than two – months of back aggregated stock data 4) NN identified a substantially greater percentage of stocks in the group with the highest explained variance than did MARS 5) data from small and midsize companies led to higher predictive accuracy than data from large size (S&P 500) companies using NN, but not MARS. The findings illustrate that the very complex interaction between insider trading data and abnormal stock returns can be systematically analyzed using non-linear techniques. Of the two assessed, NN led to comparatively more accurate predictions than did MARS.

This paper describes a new approach to find the right clustering of a dataset. We have developed a genetic algorithm to perform this task. A simple encoding scheme that yields to constant-length chromosomes is used. The objective function maximizes both the homogeneity within each cluster and the heterogeneity among clusters. Besides, the clustering genetic algorithm also finds the right number of clusters according to the Average Silhouette Width criterion. We have also developed specific genetic operators that are context-sensitive. Four examples are presented to illustrate the efficacy of the proposed method.

The induction of rules is one of the key issues of the Rough Sets Theory (RST). Generally, this problem is equivalent to finding prime implicants of a Boolean function, which is an NP-hard combinatorial problem. In practice, the NP-hardness makes solving medium-sized and large real-life problems difficult. To counteract this we propose a new algorithm, in which representation of relations between objects are represented in the form of binary vectors. The relations considered are: indiscernibility and dominance. It is an important enhancement of the classic RST approach, in which only indiscernibility was taken into account. Evaluation of the proposed algorithm in experiments with numerous real-life data sets produced satisfactory results.

In this paper, we discuss the problem of feature selection and the importance of using mutual information in evaluating the discrimination ability of feature subsets between class labels. Because of the difficulties associated with estimating the exact value of mutual information, we propose a new evaluation measure that is based on the information gain and takes into consideration the interaction between features. The proposed measure is integrated into a robust feature selection scheme and compared with the well-known mutual information feature selection (MIFS) algorithm using the problems of texture classification, speech segment classification and speaker identification.

Most of fuzzy systems use the complete combination rule set based on partitions to discover the fuzzy rules, thus often resulting in low capability of generalization and high computational complexity. To large extent, the reason originates from the fact that such fuzzy systems do not utilize the field knowledge contained in data. In this paper, based on rough set theory, a new generalized incremental rule extraction algorithm (GIREA) is presented to extract rough domain knowledge, namely, certain and possible rules. Then, fuzzy neural network FNN is used to refine the obtained rules and further produce the fuzzy rule set. Our approach and experimental results demonstrate the superiority in both rule's length and the number of fuzzy rules.