Advanced Explainable AI Framework for Dew Point Forecasting with Attention-Assisted Temporal Convolutional Networks

Patil Mounica, U. Ajay

Abstract: Forecasting of dew point temperature is an important aspect of monitoring, climate management, agriculture planning, and disaster mitigation. The moisture content of the atmosphere is determined by dew point temperature and it directly affects the weather, human comfort, crop productivity, and energy consumption. Many traditional statistical methods and popular machine learning models are unable to handle the temporal dynamics, seasonal trends and nonlinear relationships in meteorological data. While deep learning models like Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM) have shown promising forecasting accuracy, they often suffer from difficulties in capturing long-range dependency and understanding feature contributions. In response to these challenges, an Explainable AI framework to predict dew point using Attention-Based Temporal Convolutional Network (TCN) is proposed. The framework uses cutting-edge preprocessing methods such as cyclical temporal encoding, seasonal feature generation, normalization, and missing value handling to enhance data quality and representation. Explainable Artificial Intelligence using the SHAP method is used to determine which meteorological variables have the greatest influence and to gain transparency into model predictions. Moreover, a hybrid TCN-LSTM network is designed to capture the local temporal dependence and long-term sequential relationship. The attention mechanism allows the model to focus on the most relevant temporal patterns, resulting in more accurate forecasts. Experimental results show that the proposed attention-based TCN outperforms the conventional LSTM model and BiLSTM model in terms of prediction accuracy, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and coefficient of determination (R²). By combining the strengths of the convolutional approach and the recurrent memory function, the hybrid TCN-LSTM model proves to be the most effective. The results obtained have shown that the proposed framework is suitable for climate sensitive applications, environmental monitoring systems, and intelligent weather prediction platforms because it is reliable and gives very accurate results for the dew point time. The results obtained have shown that the proposed framework is reliable and gives very accurate results for the dew point time that is suitable for climate sensitive applications, environmental monitoring systems and intelligent weather prediction platforms.

Keywords: Dew Point Forecasting, Explainable Artificial Intelligence, Temporal Convolutional Network, Attention Mechanism, SHAP Analysis, Deep Learning, Climate Monitoring, LSTM, Weather Prediction, Hybrid Forecasting Models

How to Cite?: Patil Mounica, U. Ajay, "Advanced Explainable AI Framework for Dew Point Forecasting with Attention-Assisted Temporal Convolutional Networks", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 590-597, https://www.ijsr.net/getabstract.php?paperid=SR26707214952, DOI: https://dx.doi.org/10.21275/SR26707214952

Download Citation: APA | MLA | BibTeX | EndNote | RefMan