tuberculosis infection prediction based on raman and surface enhanced raman spectroscopy

Abstract: identifying the possible infections and diseases for patients hastily of time with reasonable accuracy is one of the main challenges that can be solved using machine learning [1-3]. different approaches have been discussed today, such as based on images to diagnose cancer [4]. this paper proposes a method of tuberculosis prediction based on raman and surface-enhanced raman spectroscopy. we used deep learning-based architecture models to solve the problem presented by kaewseekhao et al. [5]. the base model contains dense layers for matrix multiplication of input and predicting the outputs, the batch normalization layer that normalizes the output of the dense layer, and, dropout layers to stop the model from overfitting over train data. adam optimizer with a learning rate of 0.0001 and sparse categorical cross entropy were chosen for optimizer and loss function for the model, respectively. we train the models for 60 epochs with call back based on validation data’s accuracy to save the best model throughout the training process. two approaches were used to analyze the models. the first one was a kfold classification using ten folds, and the other separated the train and test data randomly, with the training data containing 70% of the data and the test data comprising 30%. based on the outcomes, the kfold analysis of the models shows an accuracy of 77%, and for the random-based model, the accuracy is 73%. to better understand the performance f the model, we calculate the recall too. for the kfold model, the recall was 75% and for the random based was 72%. all the codes, models, and data are available free of charge on this github repository.