The decision rules based on the rough set theory under the condition of small samples are greatly affected by the chance error of the initial data, and individual data samples are difficult to reflect the true knowledge relationship. To solve the problem of unknown reliability of rough set decision rules under the condition of small samples, concepts such as the amount of information distinction and the influence direction of attributes are proposed, and attribute weights are assigned using the Shapley value method. The influence direction of each attribute on the decision result is obtained, and the reference reliability of the decision rule is obtained to seek credible and suitable decision-making rules for engineering. Finally, the feasibility of the proposed method and the ability of discriminating the error of the data source are demonstrated through cases.