The thermal conditions of Lithium-Ion Batteries (LIBs) significantly influence their performance, making temperature regulation a vital aspect of Battery Management Systems (BMS). This paper presents an innovative technique for precise surface temperature (ST) estimation of LIBs, utilizing a Gated Recurrent Unit-Recurrent Neural Network (GRU-RNN) integrated with an Attention Mechanism. Initially, battery parameters such as voltage, current, State of Charge (SOC), and ambient temperature are inputted and dimensionally reduced via an Auto Encoder (AE) to extract salient features and minimize data redundancy. Following this, an Attention Mechanism (AM) assigns weightage to these input variables, emphasizing significant feature quantities for ST estimation. Lastly, GRU-RNN is employed to ascertain the correlation between input variables and capacity, enabling capturing of long-term dependencies amid capacity decay. This method's efficacy is confirmed through tests on LiFePO4 batteries using US06 and FUDS profiles across four distinct ambient temperatures, with a Mean Absolute Error (MAE) of less than 0.077 underpinning the accuracy of the estimations.