Postpartum depression (PPD) is a serious and under-diagnosed mental health condition affecting mothers after the birth of a child, with severe consequences for the well-being of the mother and the infant. Early identification is difficult because it is multifactorial, reported subjectively, and the interactions between biological, psychological, and social factors are complex. This research suggests an intelligent and reliable machine learning framework for early prediction of PPD using multidimensional maternal health data. The dataset contains the demographic, clinical, psychosocial, and psychological attributes obtained through structured bilingual questionnaires and Edinburgh Postnatal Depression Scale (EPDS) assessments for 64 postpartum women with 28 mixed-type features. Advanced preprocessing, feature selection, and class balancing using SMote were exploited to resolve the problem of data heterogeneity and imbalance. The proposed hybrid deep learning model combines Convolutional Neural Networks (CNN) for feature extraction at different levels, Bidirectional Long Short-Term Memory (Bi-LSTM) for modeling in time, and an attention mechanism to pay Attention to clinically relevant indicators. Model evaluation was done using a 10-fold CV based on accuracy, sensitivity, specificity, and ROC-AUC metrics. The results show the proposed model is superior to conventional approaches as it has 96.8% prediction accuracy, 96.1% sensitivity, 97.4% specificity, and ROC-AUC 0.98. Ablation analysis validates the role of each component (CNN, Bi-LSTM, Attention, feature fusion, and SMote), which has a significant contribution to the prediction performance. The study further includes explainable AI tools (SHAP and LIME) for improved interpretability for clinical applications. These results suggest that artificial intelligence-enhanced prediction of PPD can identify high-risk mothers early, facilitate early psychological interventions, and improve maternal mental health outcomes. The proposed framework provides a scalable and clinically relevant solution to supporting evidence-based maternal healthcare decision-making and could reduce undiagnosed PPD in a real-world setting.
This paper proposes a multiscale residual framework for gender classification on fingerprint images. It addresses the tough problem where most gender-related features in fingerprint ridges are subtle, and previous methods relied mainly on spatial features and failed to utilize the multi-scale features, attention, and dimension reduction. Therefore, morphological preprocessing, discrete wavelet-based multiscale feature extraction, Siamese network-based dimensionality reduction, and residual neural network classification, involving attention and non-local block, is proposed to effectively solve this problem. This framework performs morphological preprocessing on fingerprint images and decomposes the image into low-, mid-, and high-frequency channels by discrete wavelet transform; then, these channels are embedded into feature vectors by a Siamese ResNeXt-SE structure, and their cross-attention is fused to retain inter-/intra-class discriminability. Then, it applies a residual neural network with attention and a non-local block on the features for classification. On the SOCOFing dataset, which has 6,000 fingerprint images, the accuracy, precision, recall, and F1-score are 99.50%, 99.68%, 98.82%, and 0.99, respectively, and AUC is 0.99 for both genders, which achieves excellent results. Ablation studies have proven that morphological operators, wavelet decomposition, and Siamese compression do contribute to the framework individually. Experiments against two common types of noises, Gaussian and salt-and-pepper, also demonstrate the robustness against noise with trivial loss of accuracy, and comparing with prior CNN and CNN-SVM models, the method shows superiority by utilizing multiscale and frequency domain information comprehensively. However, it is constrained to use high-quality grayscale images and is not tested on adversarial robustness. The framework shows high potential for applications such as automatic fingerprint identification systems, forensics, and large biometric databases since the gender classification is effective, scalable, and discriminative.
The widespread adoption of digital learning platforms has led to increased deployment of user behavior analytics technologies to enhance learner engagement, predict their academic performance, and provide personalized learning services. Nevertheless, traditional approaches to learning raise some security issues related to the privacy violation of educational data and its possible misuse. In order to address the aforementioned challenges, this paper introduces a Federated Learning-Based Privacy-Preserving User Behavior Analytics approach (FL-PPUBA). The suggested technology allows a number of educational organizations and clients' devices to train behavioral prediction models based on data without disclosing any learner information. The algorithm uses adaptive federated averaging, encrypted parameter communication, secure aggregation, and decentralized neural network training. To conduct a computational experiment, an e-learning behavior analytics dataset including 120,000 instances was used. The framework was developed through the use of TensorFlow Federated, TensorFlow, and encrypted techniques in heterogeneous non-IID data. Evaluation of the performance was done by measuring various metrics such as accuracy, precision, recall, F1-Score, privacy preservation, and communication efficiency. The proposed FL-PPUBA framework performed better than current centralized and federated learning frameworks. Experimental findings show that the framework scored 96.4% accuracy, 95.8% precision, 95.1% recall, and 95.4% F1-Score. In addition, it obtained a privacy preservation rate of 98.1% and 91.6% communication efficiency. This indicated the performance was efficient and privacy-preserved in federated learning. Ablation studies show the contribution of both the adaptive aggregation technique and the encrypted technique in enhancing convergence stability and privacy. Based on the above analysis, it can be concluded that federated learning can provide an appropriate solution in developing privacy-preserving user behavior analytics frameworks in future e-learning applications. The next phase could consider other aspects such as blockchain technology, explainable AI, and federated intelligence at the edge level.
The problem of correctly predicting the intensity of online purchasing proves still to be very challenging due to scattered behavioural cues, extreme class imbalance and the increasing importance of trust and security in the decision-making process of consumers. Purchase prediction models currently in use have mostly been based on single behavioural measures, i.e. click rate, purchase history, etc. Single behavioural measures are not sufficient to capture the multidimensionality of purchase intention in actual e-commerce settings. To overcome these drawbacks in current paper proposes the development of Security-Conscious Composite Behavioural Intelligence (SC-CBI) model for the forecasting of online purchasing intensity at multiple levels. The proposed model introduces Composite Behavioural Indices (CBI) that introduce behavioural engagement, temporal consistency, monetary commitment and security-induced trust interactions in a low dimensional representation of user intent. In addition, a Security-Aware Balanced Classification Strategy (SABCS) will be developed in order to overcome the severe class imbalance and to preserve trust-related behavioural distributions by adaptive resampling and risk-sensitive cost optimisation. Tests on large-scale, real-world e-commerce data show that the given framework is better than traditional feature-based and imbalance-blind models. The SC-CBI model has an average accuracy of 92.7, a macro F1-score of 90.8 and a balanced recall of 90.2 which can improve up to 8-12 points compared to baselines from strong ensembles. The given framework is most appropriate for implementation in the contemporary e-commerce systems that require precision decision support in the security impaired and data imbalanced environment.
Smart agriculture uses technology to enhance farming processes, with Unmanned Aerial Vehicles (UAVs) as a crucial tool for aerial imaging, crop monitoring and precision spraying. The inability of existing research to precisely identify spray regions impedes the effective control of pests and diseases in agricultural settings. In this study, we present the development of the Clonal Flower Pollination integrated Adaptive radial deep neural network (CFPO-ARDNN), which uses data from UAV s to identify spray zones. The classifier was evaluated in agricultural settings using an RGB camera, considering five distinct croplands. To assess and test the created system, high-resolution UAV photos were employed in five distinct target fields: pea, apricot, strawberry, peach and coriander. The Joint Photographic Experts Group (JPEG) converter, RGB to hue saturation intensity (HSI), Wavelet Transform and Local Binary Pattern (LBP) for feature extraction were all used in the preparation of the video. In comparison, the CFPO-ARDNN model outperformed the others in predicting the following: strawberry (91.85%), Apricot (96.56%), Peach (95.67%), Coriander (93.47%) and pea (94.45%). The experiment’s findings demonstrate how well the suggested CFPOARDNN model works in comparison to the current approaches.
