An accelerated uptrend in IoT markets, technological breakthroughs in autonomous vehicles and artificial intelligence, which have critical latency and bandwidth constraints, have fast-tracked the adoption of edge computing. Edge data centres (EDC) are often deployed in proximity to the data sources rather than in a centralized location. Thus cooling edge data centres can be complex due to the challenges of unfavourable weather conditions and space limitations. Two-phase immersion cooling has the potential to solve these challenges, as these systems are highly energy and space-efficient, achieving overall efficiency in cooling greater than 90% and computing power as high as 100 kW/m². Further improvements in the efficiency of these systems can be achieved by holistically optimising the thermal conditions and heat transfer capabilities. Temperature prediction models facilitate the cooling system to operate closer to its maximum capacity, reducing the possibility of overcooling and improving the overall efficiency of the data centre. The proposed research addresses the cooling challenges posed by EDCs by optimising the cooling solutions in two ways − implementing a two-phase liquid immersion cooling system and via Convolution Neural Network - Long Short Term Memory based Encoder-Decoder models for temperature prediction. A novel multivariate-recursive prediction approach is employed. The time series of power and temperature of the immersed heater is the input to the model, which predicts the future temperatures as output. The cause for the temperature variation of the immersed heater is due to the changes in the power levels, which are varied through a Direct Current (DC) source using a LabVIEW program for precise control. This is similar to the server characteristics in which the heat generated by the processor raises the temperature of both the processor itself and the fluid in which the server is submerged. The proposed R-CNN-LSTM-ED thermal model predicts the temperature of the heat source with an average MAE value of 0.44, 1.42, and 2.59°C for a forecast range of 10, 30 and 60 seconds, respectively, when compared to 0.56, 1.59 and 2.78°C for R-LSTM-ED.