Marine debris is considered as a global problem and a major threat to the ecosystem and its inhabitants. To address this issue, it is important to identify these densely polluted areas inside the sea and most importantly at the seabed. In this study, we compare the traditional machine learning techniques for the classification of marine debris located deep in the sea and at the sea bed, with the aim of performing monitoring task in an automated manner. Convolutional neural network based architectures are employed for the features detection, followed by Random forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), eXtreme Gradient Boosting (XGB) and Neural Networks (NN) based classification. The comparative analysis is performed and the primary evaluation index, accuracy indicates that NN, SVM and LR using DenseNet architecture as a fixed feature extractor achieved higher test accuracy of 84%, 82% and 81% respectively. XGB classifier showed the best performance with ResNet50 network with an accuracy of 75%. Based on our findings NN and SVM are considered as the best predictors for the present deep sea marine litter dataset using DenseNet as a feature extractor.