This study aims to determine the learning activity methods (technology-based, game-based, art-based, discussion, experimental and calculation-based) that the students attended the most and the least, and reveal the predictive status of the activity characteristics (attractiveness, instructiveness and usefulness) and the target audience in the engagement of the students. A researcher-developed self-report questionnaire was implemented to 4416 students from preschool to university level in a science festival. Descriptive statistics and multinomial logistic regression analysis were used to analyze the data. The findings indicated that the highest engagement rate was for games-based activities, while the lowest was for technology-based activities. Additionally, the attractiveness, instructiveness of the activity, and the target audience predicted students' engagement in the learning activity. Consequently, increasing the attractiveness of the activity affects the probability of engaging in technology-based, game-based, and art-based activities over calculation-based. Similarly, increasing the instructiveness of the activity affects the probability of engaging in calculation-based activities over technology-based and art-based activities. The findings also showed that elementary and middle school students had similar preferences for engaging in learning activities, while high school students did not. However, the usefulness of the activity was not a predictor variable. The potential reasons for the findings were discussed and some recommendations were proposed.