Title: Finding Association Rules by Direct Estimation of Likelihood Ratios

Abstract: In this paper, we propose a cost function that corresponds to the mean square errors between estimated values and true values of conditional probability in a discrete distribution. We then obtain the values that minimize the cost function. This minimization approach can be regarded as the direct estimation of likelihood ratios because the estimation of conditional probability can be regarded as the estimation of likelihood ratio by the definition of conditional probability. When we use the estimated value as the strength of association rules for data mining, we find that it outperforms a well-used method called Apriori.