Enhancing fish freshness prediction using NasNet-LSTM

Abstract

Fish freshness is an essential feature of the fishing industry because it affects the safety and quality of the product. Precisely estimating seafood freshness is critical for consumer pleasure and waste reduction. This study presented a fish freshness prediction framework using two datasets from Kaggle, which were combined but highly imbalanced. Both upscaling (SMOTEENN) and downscaling (Random Under Sampler) methods were used to address the dataset imbalance. Neural Architecture Search Network (NasNet) and Long Short-Term Memory Networks (LSTM) models were employed to extract features from images. A feature selection technique was also applied to identify the most relevant features from the extracted features. The proposed NasNet-LSTM approach achieved impressive Matthew's correlation coefficient (MCC) and Cohen's kappa coefficient (KC) scores of 99.1%. The models were also cross-validated using a 5-fold method, resulting in MCC and KC values of 97%. Moreover, the p-value and confidence intervals of the proposed method were analyzed.