Abstract : Telecom Service Providers (TSP) face challenges in predicting Quality of Experience (QoE) as they strive to meet customer demands. Machine Learning (ML) has become a powerful tool for predicting QoE and helping TSPs exceed customer expectations. However, most existing ML algorithms for QoE prediction are batch learning-based, lacking real-time capabilities. To address this issue, a research study proposes an Incremental Discriminant based Support Vector Machine (IDSVM) model. The IDSVM model combines global and local variants using a maximum margin- based and discriminant-based approach. Unlike traditional methods, it utilizes the input space instead of the feature space to generate the variance-covariance matrix. This incremental classifier is well-suited for real-time QoE prediction, aligning with the needs of TSPs. To train the IDSVM model, the researchers utilized a dataset obtained from the Bahraini telecom regulator. This dataset, the first of its kind in deploying the novel classifier, contained technical data on the webpage accessibility of Facebook, Twitter, and Instagram. The dataset underwent several preparation phases, including data cleansing, output insertion, output balancing, and attribute selection using an autoencoder. The autoencoder generated deep features from the latent space, which served as inputs to the IDSVM model. By developing the IDSVM model and leveraging the unique dataset, this research aims to improve QoE prediction in the telecom industry, enabling TSPs to enhance customer satisfaction and maintain a positive brand image. The evaluation of the novel classifier was done using multiple performance measures, including F1 score, false positive rate, and the area under the receiver operating characteristic (ROC) curve. The training time was also considered in the evaluation. Considering the ROC measure, IDSVM outperformed the other classifiers in Twitter (0.6744) and Instagram (0.6169) datasets and scored the second lowest training time (14.66 seconds). In general, IDSVM model performance is marginally good, and it is considered a fast classifier. Therefore, it is well suited for the real-time applications.