Grassroots governance effectiveness is increasingly important as local administrations adopt data-driven decision-making to address complex social needs. Conventional evaluation methods are limited in modeling nonlinear patterns and causal relationships, which calls for intelligent evaluation frameworks. This research aims to evaluate grassroots governance effectiveness using advanced deep learning and Bayesian network methodologies. To achieve this objective, a novel Modified Sea-lion Optimizer–tuned Deep Long Short-Term Memory with Bayesian Networks (MSO-DLSTM-BN) framework is proposed. Multisource governance data are gathered from administrative records, public service performance reports, socio-economic indicators, and citizen feedback platforms across multiple grassroots units. Data pre-processing involves noise suppression using Savitzky–Golay filtering, normalization through z-score scaling, and missing value handling to ensure consistency and reliability. Feature extraction is performed using Principal Component Analysis (PCA) to reduce dimensionality while preserving dominant governance patterns. The proposed framework first structures the processed data into temporal sequences, which are learned by the DLSTM to model long-term dependencies in governance performance trends. Bayesian networks are then employed to represent probabilistic causal relationships among governance dimensions, supporting interpretable effectiveness reasoning. Governance effectiveness is evaluated across responsiveness, service equity, administrative coordination, and sustainability dimensions. The MSO-DLSTM-BN architecture developed and tested in Python has a high level of effectiveness and provides an accuracy of 0.97 and precision of 0.96 and F1-score of 0.95 with a prediction time of 2 s. It combines optimized deep temporal learning with probabilistic inference to eliminate it effectively in the evaluation of grassroots governance.
Jingyi Wang (Mon,) studied this question.