Acquired Immunodeficiency Syndrome (AIDS), caused by the human immunodeficiency virus (HIV), has been a major global public health challenge, and its prevalence has had a significant impact on the sexually active youth population, especially college students (1). According to the Joint United Nations Programme on HIV/AIDS (UNAIDS), there was an average of 3,200 new HIV infections each day among adults aged 15 and older, with approximately 30% of these infections occurring in individuals aged 15–24 in 2022 (2). Worryingly, the number of new HIV infections among college students in China has been growing sharply in recent years, increasing at an annual rate of 30–50% (3). Furthermore, the proportion of students among reported HIV cases has risen from 8.5% in 2010 to 21.7% in 2019, with 98.2% of these cases resulting from sexual transmission (4). This alarming trend indicates that college students have become a high-risk group for HIV infection. However, HIV testing rates among sexually active Chinese university students remain significantly low. For example, a study in Chongqing found that only 17.16% of adolescents aged 14–25 years with a history of sexual activity had undergone HIV testing (5). Moreover, a nationwide survey revealed substantial knowledge gaps among university students regarding HIV treatment, self-testing, and post-exposure prophylaxis (6).

During emerging adulthood, Chinese university students exhibit distinct developmental and social characteristics that profoundly influence their HIV risk and testing behaviors (7–10). Arnett’s theory identifies this stage as one of identity exploration, instability, and self-focus, with students often feeling “caught in the middle” between adolescence and full adulthood (11). In China, these characteristics are further complicated by cultural norms such as family honor and sexual conservatism. For example, while students may explore their sexual orientation and relationships in the relatively free environment of universities, they often face internal conflict due to the evolving societal acceptance of non-heterosexual identities (12). High academic pressure, future uncertainties, and strong family ties also shape their health decisions (13, 14). Moreover, the dense social networks and fluid sexual behaviors in universities increase the risk of HIV transmission (15–17). Collectively, these factors create a complex decision-making psychology around HIV testing.

Although HIV self-testing (HIVST) has emerged as an important preventive measure and early detection tool (18, 19), a key barrier contributing to this testing insufficiency remains the persistent stigma surrounding HIV/AIDS continues to prevent at-risk populations from utilizing traditional testing (20, 21) services. This stigma is particularly pronounced among adolescents, who are among the most vulnerable to HIV infection due to their earlier initiation of sexual activity and subsequent unsafe sexual behaviors, such as having multiple sex partners, engaging in casual or commercial sex, and inconsistent condom use (10, 22–24). For example, a cross-sectional study based on a large national sample of 18,000 Chinese college students (10) revealed that nearly one-third of sexually active college students initiated sexual activity before the age of 18. Meanwhile, many individuals are less inclined to undergo testing because they perceive themselves as being at low risk of contracting a Sexually Transmitted Infection (STI) (25, 26). These behaviors, combined with the social stigma surrounding HIV/AIDS, further exacerbate the challenges faced by adolescents in accessing HIV testing and prevention services. Moreover, traditional attitudes toward sexuality in Chinese society significantly impact the acceptance of HIV testing among this age group (27). For instance, the prevalence of HIV among men who have sex with men (MSM) in China is significantly higher than in the general population (28). However, HIV testing acceptance among this group remains low, partly due to the negative social attitudes and discrimination that MSM face (29). These cultural and social barriers create significant challenges for HIV prevention and testing efforts, limiting the effectiveness of traditional testing services. In this context, HIVST offers a promising solution by bypassing the barriers associated with facility-based testing, such as social stigma and geographic accessibility constraints (30–32). This approach not only addresses the unique challenges faced by adolescents and high-risk groups but also enhances overall HIV prevention strategies in the Chinese cultural context.

From a social psychological perspective, the Health Belief Model (HBM) offers a robust framework for analyzing health decision-making through six core constructs: perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action (33). These constructs collectively inform the health decision-making process. Although the HBM has been extensively applied in HIV-related behavioral research (34–36), current studies on HIVST adoption exhibit two key limitations: first, a predominant focus on the general population rather than specific behavioral subgroups (37, 38); and second, insufficient exploration of high-risk college student populations with documented unprotected sexual exposure (39, 40). This gap impedes the development of culturally competent, targeted interventions for those most vulnerable to HIV infection.

To more accurately identify the key factors influencing HIV testing and diagnosis among Chinese university students and to predict related behaviors, this study employed a random forest algorithm to rank the importance of these factors. Unlike conventional methods such as logistic regression (41), machine learning algorithms have demonstrated significant strengths in handling high-dimensional, complex, and potentially non-linear health behavior data. They are particularly well-suited for exploring behavioral prediction in the context of multifactorial interactions (42). In the field of public health, especially in HIV/AIDS research, these models have been successfully applied to predict infection risk, identify at-risk populations, optimize the targeting of interventions, and understand health service utilization patterns (42, 43). For example, He et al. utilized sentinel surveillance data of MSM individuals in Zhejiang province to predict HIV infection risk, and the random forest algorithm achieved the best performance (AUC = 0.846) (43). However, there are relatively few applications for high-risk Chinese college students.

To address these gaps, our study has three main objectives: (1) applying HBM constructs to investigate HIVST willingness among college students; (2) employing random forest modeling to identify key behavioral determinants; (3) establishing an evidence base for precision interventions targeting high-risk collegiate subgroups. This dual theoretical-empirical approach aims to enhance early HIV detection rates and inform tertiary prevention strategies tailored to the behavioral realities of university populations.

In this study, the prevalence of unprotected intercourse was found to be 15.2%, which is significantly higher than the rates reported in Jilin (8.6%) (45) and Shandong Province (5.9%) (46), suggesting potential regional disparities in sexual health literacy. This prevalence aligns with the 10–20% range documented among university students in Western countries (e.g., United States, United Kingdom) (47, 48), indicating a global challenge in young adult populations. In addition, 45.6% of participants had undergone HIVST, a rate significantly higher than that of college students in Lianyungang City (18.4%) (49) and substantially greater than the national average of 3.1% across 222 higher education institutions (50). Additionally, 703 participants (69.3%) expressed willingness to undergo HIVST within the next 6 months, surpassing the 62.5% reported in a related study in Chengdu (51). Globally, there are significant disparities in the acceptance and utilization of HIVST. For instance, in Kenya (52), the usage rate of HIVST is as high as 52%, while in Nigeria (53), only 16% of young women and 9% of young men have ever been tested for HIV. The reason for this difference could be the installation of HIVST reagent smart dispensing machines with free HIVST kits on the campus of the school (54). Global evidence indicates that HIVST devices are both convenient and efficient (55–57) and that self-testing reagents have high sensitivity and specificity (58, 59). This underscores HIVST’s potential as a scalable tool for expanding testing coverage globally, particularly in resource-limited or stigma-prone contexts (59).

Our study found that gender significantly influences willingness to self-test for HIV, with females more inclined to undergo testing than males, which contrasts with some previous findings (37, 60), where females were more inclined to undergo HIVST than males in this study. This may be because the percentage of female students participating in HIV prevention education activities in this school is higher than that of male students. Sexual behavior is the main form of HIV transmission. The rate of HIV infection among young Chinese students has risen by over 30% annually in recent years, primarily through transmission among men who have sex with me (61–63). Therefore, it is particularly urgent to increase publicity and education on HIV prevention for male students to effectively curb the spread of HIV in colleges and universities. In this survey, the higher the number of heterosexual friends and sexual partners, the more inclined to conduct HIVST. This suggests that social networks can be an important way to obtain information and influence health behaviors (64), and people are gradually recognizing the increased risk as well as the importance of health. In addition, the attenuated willingness among arts majors (OR = 0.53) warrants attention, potentially reflecting disciplinary subculture norms that merit targeted intervention (65).

To adopt healthy behaviors and change poor behaviors, the Health Belief Model is key (66). HIV awareness, including knowledge of HIV testing, has a positive impact on willingness to self-test for HIV (67, 68) and can reduce the incidence of HIV risky behaviors (69–71). Health education can increase college students’ awareness of HIV and promote the adoption of healthy behaviors to prevent HIV infection (72–74). Our HBM integration demonstrated self-efficacy (β = 0.38) and cues to action (β = 0.42) as principal behavioral drivers, while perceived barriers (β = −0.31) constituted the strongest inhibitor - findings corroborated by random forest variable importance rankings, which is consistent with the relevant studies (70, 75–78). For example, self-efficacy was the predictor of HIV testing in Brazilian youth (77), while cues to action (e.g., SMS reminders) increased clinic-based testing uptake by 30% in Kenyan university students (78). Moreover, using the random forest model for prediction, the results showed that self-efficacy and cues to action had the greatest influence on HIVST. Self-efficacy, which indicates that people who believe they are capable of performing a behavior are more likely to engage in that behavior (79). Action cues or motivation are one of the key factors in learning and a determinant of the occurrence of health behaviors (80). People with high self-efficacy are more likely to take action (81).

The above findings emphasize the urgent need to understand and address the factors that influence HIVST, especially in this high-risk population. In recent years, it has become increasingly difficult to prevent and control HIV due to the popularization of the Internet (82). Although the government and various organizations have carried out a great deal of HIV prevention and control publicity and education, the stigma, discrimination, and isolation of HIV-infected people are still relatively serious (83). High-risk groups are forced to conceal their infection status (84). In this context, the Random Forest analysis in our study validated the integration of the Health Belief Model (HBM) and provided a basis for prioritizing HIVST interventions among university students. Our findings highlight three key priorities: firstly, reducing perceived barriers emerged as the top priority, with recommendations to address financial barriers through subsidies or insurance, promote anonymous postal testing to ensure privacy, and conduct anti-stigma campaigns to correct misconceptions about HIV testing. Next, increasing self-efficacy was identified as the second priority, with strategies to enhance students’ confidence in self-testing through video demonstrations, peer education, and step-by-step goal-setting to reduce anxiety and build success experiences. Finally, optimizing action prompts was highlighted as the third priority, suggesting the placement of free testing kits in accessible locations such as health centers and dormitories, and the use of customized SMS or email reminders to stimulate testing intentions, especially around key dates like World AIDS Day. These integrated strategies aim to make HIVST more accessible, less intimidating, and ultimately increase uptake among at-risk students.

The novelty of this study is that it applies the theory of health-related behavior change to explore the factors influencing HIVST among college students at risk, and uses the random forest model to rank the importance of the influencing factors. Nevertheless, there are several limitations in this study. Firstly, the study was conducted exclusively among undergraduates at Xiangnan College, which limits the generalizability of the findings to the broader and more diverse Chinese university environment. Future research should consider replicating the study in multi-center or geographically diverse samples to enhance the external validity of the results. Secondly, the cross-sectional design limited our ability to infer causal relationships between mental constructs and test behaviors. We recommend future longitudinal or intervention-based studies to establish directional relationships. Thirdly, in this study, data on sexual behavior and HIV testing intentions were self-reported, potentially introducing social desirability or recall bias. Future research should consider validating these self-reported data using biomarkers (e.g., HIV antibody tests) and administrative records (e.g., medical HIV test logs). Fourth, during the stratified sampling process, we only stratified students by major and did not account for grade level. While major significantly influences student behavior and perceptions, grade level may also correlate with high-risk behaviors. Not considering grade level could lead to an uneven sample distribution across grades, potentially introducing selection bias. For future studies, we recommend incorporating both major and grade level in stratified sampling to enhance sample representativeness. Fifth, this study was not able to effectively recruit a sufficient number of men who have sex with men (MSM) students for the analyses, limiting our understanding of HIVST intentions in this key high-risk population. Future research should focus on this population. Finally, this study measured some HBM constructs (e.g., perceived severity and susceptibility) with binomial items, while others used Likert scales. Future research should employ consistent psychometric scales to improve result reliability.

In conclusion, our dual-model approach identifies modifiable predictors (self-efficacy, cues to action) and high-risk subgroups (arts majors, socially active males) for targeted intervention. The integration of accessible testing infrastructure with HBM-informed education programs, particularly addressing male-specific prevention gaps and disciplinary subcultures, could optimize campus HIV control. Subsequent research should explore hybrid machine learning architectures to capture complex behavioral interactions, ultimately advancing precision public health strategies.