This paper explores methodologies to enhance the integration of a green supply chain circular economy within smart cities by incorporating machine learning technology. To refine the precision and effectiveness of the prediction model, the gravitational algorithm is introduced to optimize parameter selection in the support vector machine model. A nationwide prediction model for green supply chain economic development efficiency is meticulously constructed by leveraging public economic, environmental, and demographic data. A comprehensive empirical analysis follows, revealing a noteworthy reduction in mean squared error and root mean squared error with increasing iterations, reaching a minimum of 0.007 and 0.103, respectively—figures that are the lowest among all considered machine learning models. Moreover, the mean absolute percentage error value is remarkably low at 0.0923. The data illustrate a gradual decline in average prediction error and standard deviation throughout the model optimization process, indicative of both model convergence and heightened prediction accuracy. These results underscore the significant potential of machine learning technology in optimizing supply chain and circular economy management. The paper provides valuable insights for decision-makers and researchers navigating the landscape of sustainable development.

In recent years, deep reinforcement learning (DRL) models have been successfully utilised to solve various classification problems. However, these models have never been applied to customer credit scoring in peer-to-peer (P2P) lending. Moreover, the imbalanced class distribution in experience replay, which may affect the performance of DRL models, has rarely been considered. Therefore, this article proposes a novel DRL model, namely a deep Q-network based on a balanced stratified prioritized experience replay (DQN-BSPER) model, for customer credit scoring in P2P lending. Firstly, customer credit scoring is formulated as a discrete-time finite-Markov decision process. Subsequently, a balanced stratified prioritized experience replay technology is presented to optimize the loss function of the deep Q-network model. This technology can not only balance the numbers of minority and majority experience samples in the mini-batch by using stratified sampling technology but also select more important experience samples for replay based on the priority principle. To verify the model performance, four evaluation measures are introduced for the empirical analysis of two real-world customer credit scoring datasets in P2P lending. The experimental results show that the DQN-BSPER model can outperform four benchmark DRL models and seven traditional benchmark classification models. In addition, the DQN-BSPER model with a discount factor γ of 0.1 has excellent credit scoring performance.

For real-world customer classification data, data structures are often highly uncertain. The constructed models may not match the data structure characteristics, which can lead to unsatisfactory classification performance. Further, the class distribution characteristics of data sets are usually highly imbalanced, which can also weaken the classification performance. To solve the above problems, we construct a complex-valued group method of data handling (CGMDH) neural network-based data characteristic-driven adaptive decision support system. First, we introduce the linearly separable discriminant method to analyze the data structure characteristics. Second, we extend the circular linear CGMDH neural network model and propose a circular quadratic nonlinear CGMDH (QCGMDH) neural network model. Finally, according to the data structure characteristics and resampling technique, we adaptively select and train the most appropriate CGMDH neural network model from two types of CGMDH. To analyze the effectiveness of the constructed system, the experimental results of 16 real-valued classification data sets show both linearly separable discrimination and random oversampling technology can help to improve the classification performance. Further, to verify its customer classification performance, we conduct an empirical analysis on 14 real-valued customer classification data sets and find that its customer classification performance is significantly better than that of the other nine models and comparable to that of the circular QCGMDH neural network model.

The fractional-order grey prediction model is widely recognized for its performance in time series prediction tasks with small sample characteristics. However, its parameter-estimation method, namely the least squares method, limits the predictive performance of the model and requires time to address the ill-conditioning of the system. To address these issues, this paper proposes a novel parameter-acquisition method treating structural parameters as hyperparameters, obtained through the marine predators optimization algorithm. The experimental analysis on three datasets validate the effectiveness of the method proposed in this paper.