The quality (Q) factor and the mode volume are the two most important properties of optical resonances realized by Photonic Crystal (PC) nanocavities. A higher Q-factor means a stronger ability of the cavity to confine optical energy inside of itself and a smaller mode volume will lead to smaller device foot- print and higher degree of integration. The traditional method to calculate the Q factor and mode volume of a PC nanocavity with a fixed design is to verify it in a simulation software. To improve the efficiency of the validation procedure, a Convolutional Neural Network (CNN) based approach has been designed to predict these two target values. In this paper, an overview of existing predictive CNN models is provided, and the hyperparameter optimization used to improve the accuracy and correlation of predictions by the model is shown. All in all, the authors propose a possible optimization strategy for the current CNN model applied in photonics forward predictions and also provide a general framework for the hyperparameter optimization of CNNs with non-image inputs and multiple outputs.