Abstract: Chilling injury, a major meteorological hazard, remarkably constrains the stable and high‐yield production of rice and thus directly affects the resilience of regional agricultural systems.Although the overall frequency of chilling events has decreased as a consequence of global warming, the risk of extreme low‐temperature events has continued owing to an increase in climate variability in high‐latitude regions. Therefore, an accurate assessment of a chilling injury, especially the delayed type, is essential for ensuring the stability and sustainability of rice production, forming the basis for developing effective prevention and mitigation strategies. However, existing chilling injury indices are mostly limited by reliance on singular indicators, static baseline thresholds and homogeneous criteria for the classification of severity. To overcome these limitations, this study aimed to introduce an evaluation index for delayed‐type chilling injury in rice by integrating data‐driven techniques with hyperparameter optimisation to improve the accuracy and adaptability of assessments. This method dynamically calculates daily anomalies using heat indices and 30‐year moving averages. Using the cumulative negative anomaly and the maximum number of consecutive days having negative anomalies as key variables, the entropy weight method is employed to quantify the relative importance of each disaster‐inducing factor. Accordingly, a comprehensive cold intensity index (CCII) is constructed. To determine the criteria for classifying the severity of chilling injury, coarse‐ and fine‐grained grid search techniques are used along with cross‐validation. This method enables the precise classification of chilling injury severity by establishing differentiated grading thresholds, which were used to assess chilling events across the three northeastern provinces of China. Furthermore, this method was used for the spatiotemporal analysis of chilling injury frequency. The study results revealed that the CCII‐based classification agreed well with historical disaster records and achieved an accuracy rate of 80.77%. The index accurately identified representative years in which remarkable chilling injury occurred (i.e. 1969, 1992 and 2009), as well as years with negligible impact (e.g. 2007), and was consistent with historical documentation. The spatiotemporal analysis of chilling frequency revealed that severe chilling events were mainly concentrated in northern Heilongjiang and eastern Jilin. From the 1960s to the 2010s, the overall frequency of rice chilling injury exhibited a declining trend, with the 1970s symbolised as the peak period. Although the frequency of severe chilling events has markedly decreased, localised outbreaks sometimes occur in certain regions, indicating that the potential risk of extreme chilling events remains critical. The delayed‐type chilling injury index developed in this study offers a precise and quantitative means for assessing the frequency, intensity and duration of low‐temperature damage to rice. This index overcomes the limitations of conventional indices and provides a robust new approach for accurately assessing chilling injury. Moreover, it offers scientific support for decision‐making about rice production across Northeast China.