Air pollution in cities has a massive impact on human health, and an increase in fine particulate matter (PM${}_{2.5}$) concentrations is the main reason for air pollution. Due to the chaotic and intrinsic complexities of PM${}_{2.5}$ concentration time series, it is difficult to utilize traditional approaches to extract useful information from these data. Therefore, a neural model with a dendritic mechanism trained via the states of matter search algorithm (SDNN) is employed to conduct daily PM${}_{2.5}$ concentration forecasting. Primarily, the time delay and embedding dimensions are calculated via the mutual information-based method and false nearest neighbours approach to train the data, respectively. Then, the phase space reconstruction is performed to map the PM${}_{2.5}$ concentration time series into a high-dimensional space based on the obtained time delay and embedding dimensions. Finally, the SDNN is employed to forecast the PM${}_{2.5}$ concentration. The effectiveness of this approach is verified through extensive experimental evaluations, which collect six real-world datasets from recent years. To the best of our knowledge, this study is the first attempt to utilize a dendritic neural model to perform real-world air quality forecasting. The extensive experimental results demonstrate that the SDNN offers very competitive performance relative to the latest prediction techniques.

With the rapid development of artificial neural networks, recent studies have shown that dendrites play a vital role in neural computations. In this study, we propose a dendritic neuron model called the approximate logic dendritic neuron model (ALDNM) to solve classification problems. The ALDNM can be divided into four layers: the synaptic layer, the dendritic layer, the membrane layer, and the soma body. Considering the limitation of the back-propagation (BP) algorithm, we employ a heuristic optimization called the social learning particle swarm optimization algorithm (SL-PSO) to train the ALDNM. In order to investigate the effectiveness of SL-PSO for training the ALDNM, we compare this training method with BP and four other typical heuristic optimization methods. Moreover, the proposed ALDNM is also compared with seven classifiers to verify its performance. The experimental results and statistical analysis on four classification problems indicate that the proposed ALDNM trained by SL-PSO can provide a competitive performance for solving the classification problems. It is worth emphasizing that the structure of the trained ALDNM can be greatly simplified owing to the unique pruning operations. Furthermore, the simplified ALDNM for a specific problem can be converted into a corresponding logic circuit classifier for a fast classification.

The conventional dendritic neuron model (DNM) is a single-neuron model inspired by biological dendritic neurons that has been applied successfully in various fields. However, an increasing number of input features results in inefficient learning and gradient vanishing problems in the DNM. Thus, the DNM struggles to handle more complex tasks, including multiclass classification and multivariate time-series forecasting problems. In this study, we extended the conventional DNM to overcome these limitations. In the proposed dendritic neural network (DNN), the flexibility of both synapses and dendritic branches is considered and formulated, which can improve the model's nonlinear capabilities on high-dimensional problems. Then, multiple output layers are stacked to accommodate the various loss functions of complex tasks, and a dropout mechanism is implemented to realize a better balance between the underfitting and overfitting problems, which enhances the network's generalizability. The performance and computational efficiency of the proposed DNN compared to state-of-the-art machine learning algorithms were verified on 10 multiclass classification and 2 high-dimensional binary classification datasets. The experimental results demonstrate that the proposed DNN is a promising and practical neural network architecture.

Student performance prediction was deployed to predict learning performance to identify at-risk students and provide interventions for them. However, prediction models should also consider external factors along with learning activities, such as course duration. Thus, we aim to distinguish the difference factor between the time dimension (duration of the course) and the feature dimension (students’ learning activities) by attention weights to provide helpful information and improve predictions of student performance. In this study, we introduce Attention-Based Artificial Neural Network (Attn-ANN), a novel model in educational data mining. The Attn-ANN combines attention weighting on the time and feature dimensions to examine the significance of lectures and learning activities and makes predictions by visualizing attention weight. We found that the Attn-ANN had a better area under the curve scores than conventional algorithms, and the attention mechanism allowed models to focus on input selectively. Incorporating the attention weighting of both the time and feature dimensions improved the prediction performance in an ablation study. Finally, we investigated and analyzed the model’s decision, finding that the Attn-ANN may be able to create synergy in real-world scenarios between the Attn-ANN’s predictions and instructors’ expertise, which underscores a novel contribution to engineering applications for interventions for at-risk students.