As a neurodegenerative disorder, Parkinson’s disease (PD) affects the nerve cells of the human brain. Early detection and treatment can help to relieve the symptoms of PD. Recent PD studies have extracted the features from vocal disorders as a harbinger for PD detection, as patients face vocal changes and impairments at the early stages of PD. In this study, two hybrid models based on a Support Vector Machine (SVM) integrating with a Principal Component Analysis (PCA) and a Sparse Autoencoder (SAE) are proposed to detect PD patients based on their vocal features. The first model extracted and reduced the principal components of vocal features based on the explained variance of each feature using PCA. For the first time, the second model used a novel Deep Neural Network (DNN) of an SAE, consisting of multiple hidden layers with L1 regularization to compress the vocal features into lower-dimensional latent space. In both models, reduced features were fed into the SVM as inputs, which performed classification by learning hyperplanes, along with projecting the data into a higher dimension. An F1-score, a Mathews Correlation Coefficient (MCC), and a Precision-Recall curve were used, along with accuracy to evaluate the proposed models due to highly imbalanced data. With its highest accuracy of 0.935, F1-score of 0.951, and MCC value of 0.788, the probing results show that the proposed model of the SAE-SVM surpassed not only the former model of the PCA-SVM and other standard models including Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbor (KNN), and Random Forest (RF), but also surpassed two recent studies using the same dataset. Oversampling and balancing the dataset with SMOTE boosted the performance of the models.

The widespread use of online social media platforms has amplified the importance of efficient hate speech detection, especially in low-resource languages like Bengali. While traditional machine learning approaches show promise, deep learning is more effective in capturing the nuanced context of hate speech. Current challenges include a lack of diverse datasets and models capable of context-sensitive detection. To address these, we introduce HateCorpBN-XL, the largest labeled Bengali hate speech dataset to date, containing 65,251 comments across five categories: political (PoHS), religious (ReHS), misogynistic (MisoHS), slander (SlaHS), and xenophobic (XenHS). We also propose HateBertBN, a hybrid transformer-based model combining BanglaBERT embeddings with three neural network fusion strategies using CNN, LSTM, and MLP. We evaluate our approach on two tasks, Task-1: detecting hate speech in Bengali text classifying it as hateful or non-hateful and Task-2: categorizing hateful content into five distinct classes. For Task-1, all HateBertBN variants outperformed current transformer models, achieving an accuracy of 0.92 and a weighted F1-score of 0.92. In Task-2, the HateBertBN-MLP and HateBertBN-CNN variants achieved a notable 0.90 accuracy and weighted F1-score of 0.90, surpassing M-BERT, Distil-M-BERT, BanglaBERT, and XLM-R-Base. Although HateBertBN-LSTM performed slightly lower overall, it achieved strong F1-scores in the ReHS (0.93) and XenHS (1.00) categories. Overall, our hybrid model outperforms state-of-the-art approaches in both tasks, demonstrating its effectiveness and robustness.