Objective This research aims to develop a comprehensive environmental health risk prediction system for elderly populations in cold regions based on a Bayesian probability framework. The system is designed to quantitatively evaluate the effects of different outdoor environmental exposures on physiological and psychological indicators of elderly individuals, thereby providing evidence-based decision support for elderly health management and elderly-oriented environmental design. The research addresses the unique challenges faced by the elderly in cold regions, where prolonged low temperatures significantly impact cardiovascular health and outdoor activity patterns, creating special health management challenges for this vulnerable population. By incorporating individual difference parameters and environmental characteristic metrics into a predictive framework, the research seeks to bridge the gap between theoretical knowledge and practical applications in elderly-oriented landscape design.

Methods The research employs a multi-stage methodological approach combining field experimentation, mathematical modeling, and application development. Health indicators of elderly subjects (n = 345, aged 60 − 70) are collected in three distinct outdoor environments (activity area, greenway area, and street area) in a community in Shenyang, China. Data collection was conducted during November − December of 2023 and 2024, with outdoor temperatures ranging from 4°C to 8°C. Environmental parameters are standardized through a two-tier framework quantifying spatial openness (δ_open) and green coverage (δ_green) relative to reference standards. Individual sensitivity parameters are established incorporating gender differences, with sensitivity coefficients (η) and regulatory factors (γ) calculated based on physiological responses. A systematic testing is conducted following a standardized protocol consisting of preparation, environmental exposure, and recovery assessment phases. Physiological indicators include systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse pressure (PP), which are measured using an Omron HEM-7136 electronic sphygmomanometer. Psychological variables are assessed using validated Chinese versions of the Profile of Mood States (POMS) and Restoration Outcome Scale (ROS) with Cronbach’s α coefficients of 0.86 and 0.82 respectively. Based on the collected data, a Bayesian probability model is constructed that transforms traditional Bayesian components into environment-specific parameters: Prior probabilities become baseline blood pressure values, likelihood functions become environmental adjustment effects, and posterior distributions become predictive blood pressure values with confidence intervals. The Artificial Intelligence-Integrated Development Environment (AI-IDE) platform is utilized to transform the theoretical framework into a practical application. The development process employs an iterative evolution approach, converting the Bayesian probability framework into executable code through natural language processing capabilities of the AI-IDE platform. The resulting mobile-based environmental health risk prediction system is featured by elderly-oriented interface design with optimized interaction elements for elderly users.

Results The research identifies significant gender-based differences in environmental sensitivity, with elderly females demonstrating markedly higher sensitivity coefficients compared to males (0.85±0.04 vs. 0.72±0.05) and greater regulatory factors (1.24±0.07 vs. 0.86±0.05). These differences are manifested in physiological responses, with female subjects exhibiting larger blood pressure fluctuations during environmental transitions (8.76±2.31 mmHg vs. 5.24±1.87 mmHg). Among the three outdoor environments, the greenway area produces the most positive health effects, characterized by a mean decrease in systolic blood pressure of 2.7±1.8 mmHg from baseline and improvements in psychological indicators (POMS scores decrease by 2.6±0.9, while ROS scores increase by 0.53±0.12). Conversely, the street area induces negative effects, with SBP increasing by 7.8±2.4 mmHg on average and negative mood indicators rising. The activity area demonstrates intermediate effects with non-significant SBP changes (±1.5 mmHg) and slight mood improvements. The prediction model demonstrates excellent performance metrics across validation testing. The system performs best in predicting responses in the activity area (SBP mean root error: 4.8 mmHg; accuracy rate: 91.2%), with slightly higher error rates in street area, where the accuracy rate is still maintained above 88.5%. Five-fold cross-validation confirms model stability with a CV coefficient of 0.092. Overall model fit achieves an R² value of 0.87, with prediction interval coverage reaching 93.8%, demonstrating strong explanatory power and reliability. Key health indicators (SBP, POMS, and ROS) all show significant linear relationships between predicted and actual values. The mobile terminal implementation features age-appropriate design elements including large-sized touch control elements (30px × 30px with a minimum spacing of 12mm), high contrast visual feedback, 18px font size, and a three-tiered risk visualization framework using color-coding (green − orange − red) to enhance information accessibility for elderly users.

Conclusion The prediction system based on the Bayesian probability framework successfully achieves accurate assessment of environmental health risks for elderly individuals in cold regions. The integration of individual difference parameterization through sensitivity coefficients and regulatory factors, combined with a multi-level linkage framework connecting environmental features to health outcomes, significantly improves prediction accuracy for risk probability. The system effectively addresses the common challenges of small-sample health research by leveraging Bayesian approaches to handle uncertainty in parameter distributions, providing robust predictions despite limited training data. The application of AI-IDE platform notably accelerates the transformation process from research findings to practical applications, establishing a seamless bridge between academic knowledge and implementable tools. This approach substantially lowers technical barriers for cross-disciplinary applications by converting research requirements and model logic into structured code through natural language processing. The system provides quantitative indicators and scientific foundations for optimizing elderly-oriented landscape environments in cold regions, including optimal spatial openness range (0.65 − 0.80), recommended green coverage threshold (0.82 − 0.88), and gender-specific environmental transition zone designs. These evidence-based design parameters offer practical guidance for creating outdoor environments that enhance physiological and psychological well-being of elderly populations in cold regions, ultimately supporting healthy aging in place.