Deep Learning-Driven Visual Analytics Framework for Real-Time Data Interpretation and Predictive Decision Making

Abstract

The rapid growth of real-time data generated from environmental sensors, social media platforms, and IoT-enabled systems has created significant challenges in extracting actionable insights for timely decision-making. This study introduces a novel approach called Deep Learning-Driven Visual Analytics Framework for Real-Time Data Interpretation and Predictive Decision Making that combines multiple data sources to enhance the precision of data interpretation and decision-making. The framework combines the Air Quality Dataset and Sentiment140 Dataset, which are used to merge environmental measurements with sentiment information from the public. A complete pre-processing pipeline is used, which includes missing value imputation, normalization, cleaning text, tokenisation, lemmatisation and creation of BERT embeddings. Relevant features are extracted from both sets of data, and combined into a single multimodal representation. Then a hybrid CNN-BiLSTM network is applied to capture spatial and temporal dependences for accurately prediction and classification. The experimental results show that the model has a better prediction performance for air quality, with an MAE of 0.041, RMSE of 0.072 and R² score of 0.97, and a more accurate results for sentiment classification with 98.83% accuracy and an F1 score of 98.17%. The proposed framework improves the real-time monitoring, predictive intelligence, and visual decision making for various environmental, healthcare, industrial and smart city applications.