A Developed Tunable Model against Prebuild Applications to Classify Emotions

Building a classification model on images based on convolutional neural network concerns several factors. Different solutions were explored and presented, which include the implementation of neural architecture search to automate the selection of a best architecture taking into consideration a certain dataset. To provide relevant materials in the development of a tunable model and utilization of the available Keras Tuner hyper models, this study was conducted. FASS, the developed tunable model to classify emotions trained on CK+ dataset was able to generate a model with 93.75% accuracy on test data. The performance of FACE-FASS, the developed convolutional neural network classification model, was higher than the performance of the model built from using the HyperResNet or HyperXception prebuild applications trained on the same dataset. Despite the higher performance, further study is recommended for a possible increase in its accuracy, as well as its application on other image datasets.