Weather prediction is an appealing but demanding endeavor because of its substantial effects on human existence and the complex dynamics of atmospheric movement. Importance of weather forecasting is huge in daily life activities, business, agriculture etc. so; scholarly genre is taking great interest in this field. In weather prediction large amounts of data have been collected from multiple sources such as satellites, weather stations, radar and historical records, is complex task to process. Machine learning and deep learning approaches that rely on a huge amount of data with quickly and accurately become more popular. Numerous technique focus only temporal pattern of meteorological data, ignoring the correlations between multiple variables at various geographical locations. In this paper Chaotic Logistic Map Based Grey Wolf Optimization (CLMGWO) determine appropriate climate factor for each geographical location and Attention based Gated Recurrent Unit (AttGRU) provide a precise prediction of feature correlation with many parameter and station across temporal time stamp. Proposed method AttGRU_CLMGWO resolve the problem of feature selection with successfully capture concealed spatial interconnections and a wide range of enduring weather patterns. Finally AttGRU implemented with Root Mean Square Propagation (RMSProp) Optimizer to evaluation of mean square error. This error is used to recalibrate the weight and bais in order to get improved result. AttGRU_CLMGWO model is comprised with Graph Neural Network (GNN), Bayesian Multi-head Attention Encoder-Decoder Neural Network (BMAE-Net) and Convolutional Neural Networks (CNN). Proposed model AttGRU_CLMGWO has implemented in python using Jena Climate dataset and predict temperature and humidity by concurrently acquire data for crucial time stamp and weather forecasting. The results yielding from the AttGRU_CLMGWO is MSE - 1.3, an MAE - 0.41, and a MAPE - 0.2.