To reduce effectively the reading anxiety of learners while reading English articles, a C4.5 decision tree, a widely used data mining technique, was used to develop a personalized reading anxiety prediction model (PRAPM) based on individual learners' reading annotation behavior in a collaborative digital reading annotation system (CDRAS). In addition to forecasting immediately the reading anxiety levels of learners, the proposed PRAPM can be used to identify the key factors that cause reading anxiety based on the fired prediction rules determined by the developed decision tree. By understanding these key factors that cause reading anxiety, instructors can apply reading strategies to reduce reading anxiety, thus promoting English-language reading performance. To assess whether the proposed PRAPM can assist instructors in reducing the reading anxiety of learners, this study applies the quasi-experimental method to compare the learning performance of three learning groups, which are, respectively, supported by a CDRAS with individual annotations, collaborative annotations, and collaborative annotations with online instructor's support to reduce reading anxiety by the proposed PRAPM. The instructional experiment was conducted on Grade 7 students at Taipei Municipal Wan-Fang high school. Experimental results indicate that the average correct prediction rate of the proposed PRAPM in identifying the reading anxiety levels of learners was as high as 70%. Moreover, analytical results show that the collaborative annotation with online instructor's support for reducing reading anxiety by the proposed PRAPM indeed helps learners reduce reading anxiety, particularly for the male learners, showing that gender difference exists. Furthermore, based on online instructor's support for reducing reading anxiety by the proposed PRAPM, the correlation analysis also shows that the online instructor's interaction with the male learners is significantly correlated with the reading anxiety reduction. Furthermore, English-language learning performance of the three learners groups, which were given a CDRAS with different learning mechanisms, was significantly promoted.