marine pathogens classification based on raman spectra using deep learning

Machine learning and deep learning approaches have been an important impact in different fields of chemistry, especially analytical chemistry, and it has lots of potential for the future. using raman spectra to identify and predict specific tasks has been a subject of interest [1]. accurate determination and prediction of pathogens in the environment have been gathered in different researches to develop fast and accurate models to predict this specific property [2]. therefore, we developed a deep learning model based on raman spectra data that was gathered and published by yu et al [2] to generate a more accurate model. the model architecture consists of a dense layer with an input shape of 1200, a batch normalization layer, and a dropback layer. these layers were repeated three times, and at the end, a single dense layer with softmax activation and output of eight was put to predict the class of the marine pathogens based on their raman spectra data. the input data contains of 1200 rows. our model outperformed the previous model by achieving the average accuracy for 5fold cross-validation of 99.91% and the recall of 99.9. for the random split model, data is separated into two different datasets. 70% on the training dataset and 30% on the test dataset; in this situation, the proposed model achieved 100% accuracy and the recall of 100% on the test dataset. for both of these process, a checkpoint method was used to save the best model based on its performance on the test dataset. the deep learning models were generated using tensorflow version two. all the codes, data, and trained models are available free of charge at this github repository.