The Burns Depression Checklist (BDC) measures depression severity by adding the scores of 25 symptom items. However, the total score does not show which items contribute most to the severity category. This study presents an interpretable machine learning pipeline to rank predictive BDC items. We analyzed responses from 654 university students in Bangladesh. We first examined the reliability and relationships of the BDC items. We then tested seven machine learning classifiers. We also applied four feature selection methods—RFE, SKB, UFS, and MIFS—and two explanation methods, SHAP and LIME. The BDC showed strong internal consistency, with a Cronbach’s alpha of 0.91. The full-item stacking model achieved 90.84% test accuracy, 86.03% balanced accuracy, and 82.60% macro-F1. We combined the six item rankings on one common scale. Loss of motivation received the highest rank. Loss of interest in work, loneliness, discouragement, hopelessness, and reduced social interest also ranked highly. These items represent reduced motivation, social withdrawal, negative mood, and negative self-view. The main contribution of this study is an item-level ranking that combines evidence from several analytical methods. However, the target categories come directly from the same 25 BDC items. Therefore, the results show how well the models reconstruct BDC severity categories, not how well they diagnose clinical depression. The rankings also do not show clinical urgency. Any positive response to a suicide related item requires separate professional assessment.
Brain tumors refer to the uncontrolled growth of cells in the human brain, posing a significant risk to human health. Therefore, early detection of these tumors are crucial. However, existing feature extraction approaches often fail to capture the intricate characteristics of tumors, leading to suboptimal classification performance. To address this limitation, a Multi-scale Convolutional Neural Network and Gated Channel Transformation unit-based DenseNet-169 model (MC-GCT-DenseNet-169) is proposed for efficient brain tumor classification. The multi-scale convolutional neural network captures the nonlinear relationships between texture, shape, and intensity in magnetic resonance images (MRI). The GCT function in the DenseNet-169 model enhances the ability to learn subtle variations in MRI such as tumor texture and boundary features, which are essential for accurate classification. Experimental results show that the MC-GCT-DenseNet-169 method attains accuracy of 98.97% on the Figshare dataset, surpassing previous approaches such as Cross-prior Bayesian attention mechanism with a Correlated Inception and Residual Learning Network (CB-CIRL Net).
Irrigation management in tropical fruit orchards is most difficult during the tran- sition from the dry to the wet season, when atmospheric demand is high and rain- fall is erratic. This study presents a web-based decision-support framework that converts hourly meteorological records into irrigation requirements and evalu- ates how they shift under climate perturbation. Vapor pressure deficit (VPD) and hourly reference evapotranspiration (ETo) are computed with the FAO-56 Penman-Monteith equation, scaled by stage-specific crop coefficients and con- verted to a per-tree volume using canopy radius. The implementation reproduces the worked hourly values of FAO-56 Example 19 exactly. For a representative day in Chanthaburi, Thailand, the net requirement for durian rose from 3.53 to 7.05 liters per day for a tree of 1.0 m canopy radius under a perturbation of +2 ◦C and −20% relative humidity. Because this result depends on how hu- midity is assumed to respond to warming, the calculation was repeated under three assumptions which gave increases in mean VPD of between 12.1 and 69.9 percent and in ETo of between 4.8 and 20.4 percent. The direction of the shift is therefore robust while its magnitude is assumption-dependent and irrigation requirements are reported as a range rather than a single value.
Heat pumps play a crucial role in reducing carbon emissions in residential heating; however, their electricity consumption varies significantly across households, making large-scale performance assessment challenging. Existing data-driven approaches often rely on fully supervised learning and require extensive labeled data, which are costly to obtain through on-site inspections. This paper investigates whether routinely collected smart meter and weather data can be leveraged to automatically classify household heat pump electricity consumption into three categories: low, normal, and rather high. We propose a self-monitoring transformer framework that combines self-supervised pretraining and supervised fine-tuning for time series classification. A transformer encoder is first pre-trained on multi-day electricity consumption windows from largely unlabeled households using a reconstruction-based objective. The learned representations are then fine-tuned with a lightweight fully connected classifier on a limited labeled subset of the HEAPO dataset. Experimental results demonstrate that the proposed approach consistently outperforms state-of-the-art baselines, including RNN-, CNN-, and transformer-based time series models, achieving approximately 93% classification accuracy, an F1 score above 0.9, and a recovery rate close to 91%. These results indicate that self-supervised representation learning on smart meter data provides an effective and scalable solution for monitoring heat pump electricity consumption and supporting inspection planning.
This paper presents an integrated computational framework combining parameter estimation and optimal control for sustainable mangrove forest management. We begin by addressing uncertainty in model parameters using a data-driven solution. A least-squares optimization problem is solved using the Levenberg-Marquardt algorithm to recursively estimate parameters in the dynamic mangrove ecosystem model. Later, we formulate an optimal control problem to drive mangrove populations toward ecological steady states as rapidly and efficiently as possible. To this end, we define the Hamiltonian function and derive the necessary optimality conditions to design optimal control policies. Our key contribution lies in unifying parameter estimation and control synthesis into a single, implementable algorithm tailored for mangrove dynamics. Numerical simulations demonstrate the algorithm’s efficacy: parameter estimates achieve high accuracy, as evidenced by low mean squared error, while the optimal control policy minimizes a defined performance index, ensuring ecological and operational efficiency. This integrated approach offers decision-makers a data-driven, mathematically rigorous tool to guide restoration and conservation efforts. By bridging estimation and control, our methodology enhances adaptive management of mangrove forests, supporting long-term ecological resilience and sustainability in dynamic coastal environments.
