The global prevalence of depression is 4.4 % (Organization, 2017), and it remains a major cause of global health burden (GBD 2019 Mental Disorders Collaborators, 2022; Murray, 2024; Yang et al., 2013). Anxiety is the most common mental health disorder and the sixth leading cause of disability globally (Organization, 2017). In the past decade or so, both conditions have shown increasingly prevalence among university students worldwide (Ibrahim et al., 2013). This population faces distinct stressors, including academic pressures, social adaptation challenges, and identity development, which increase their susceptibility to mental health problems (Brown, 2018). This not only affects students' learning and quality of life, but may also have long-term negative consequences for their future development. Therefore, early screening and intervention for mental health issues in university students are of significant public health and societal importance (Duffy, 2023).

Although widely validated, traditional clinical diagnostic scales rely on subjective self-report and may yield inconsistent results (Association, 2013). Biological and physiological diagnostic tools, such as cortisol assays and neuroimaging, offer more objective alternatives (Gold et al., 2015; Lai, 2019; Zang et al., 2022), but these methods are costly, time-intensive, and difficult to implement at scale. With the growing prevalence of mental health disorders and increasing pressure on healthcare systems, the need for scalable, low-cost, and objective screening tools is urgent.

Physical activity is crucial for both physical and mental health (Bize et al., 2007; Bull et al., 2020; Miles, 2007; Varma et al., 2014; Warburton et al., 2006). Research shows that regular physical activity can prevent or alleviate depression symptoms, with even relatively small amounts providing measurable protective benefits (Mammen and Faulkner, 2013; Teychenne et al., 2008), and high-intensity activity is associated with reduced depression risk across all age groups (Schuch et al., 2018). For example, a previous study indicated that meeting the recommended 2.5 h of brisk walking per week was estimated to reduce depression risk by 11.5 % (Pearce et al., 2022). Similarly, systematic evidence supports the role of physical activity in reducing anxiety symptoms (McDowell et al., 2019). However, most prior studies rely on self-reported physical activity, which is susceptible to recall and reporting bias (Jonsdottir et al., 2010; Prince et al., 2008; Rodríguez-Romo et al., 2023).

Daily step counts are a low-cost and objective source of behavioral data that can reflect individuals' physical activity levels, lifestyle regularity, and daily routines. With the development of sensor technology, step counts can be accurately measured, proving an objective proxy for physical activity (Bizzozero-Peroni et al., 2024; Pearce et al., 2022). Evidences supporting the mental health benefits of walking are growing (Dang et al., 2023; Fennell et al., 2022; Kelly et al., 2018; Washburn and Ihm, 2021; Wolf et al., 2021). Previous studies have found that higher daily step counts are associated with lower depressive symptom severity (Hammett et al., 2023; Lee et al., 2014; Ludwig et al., 2018; Recchia et al., 2023). However, these studies primarily relied on basic statistical metrics of step counts (e.g., average value), and have not fully explored step count–derived indicators that capture interpretable behavioral characteristics such as lifestyle periodicity and regularity. Furthermore, previous research has predominantly utilized traditional pedometers (McKercher et al., 2009), accelerometers (Lee et al., 2014) and other wearable monitoring devices (Ainsworth et al., 2015), which have drawbacks such as poor user compliance and high collection costs.

To address these limitations, we developed a low-cost and convenient app-based step count collection system and explored the associations between step counts and mental health status among university students. In addition to extracting traditional statistical features, we innovatively treated step counts as time-series signals to extract frequency domain and non-linear features, reflecting the periodicity and regularity of lifestyle patterns. We further introduced a novel Weekday–Weekend Activity Fluctuation Index (WWAF) to quantify deviations in daily routine patterns. Finally, we performed machine learning classification models to predict depression/anxiety status using extracted features.