Kmer - Fully Connected Neural Network

This repository includes the results of running the FCNN on 3mer, 4mer and 5mer on the following taxonomic levels: PHYLUM, CLASS, ORDER, FAMILY and GENUS. There are 4 files, corresponding to the result of training, validating and testing with the 4 datasets created in the Data Preparation phase (5% cut-off with ambiguous nucleotides, 5% cut-off without ambiguous nucleotides, 1% cut-off with ambiguous nucleotides, 1% cut-off without ambiguous nucleotides).
To find the optimal balance between running time and the F1 score, the 3mer model was trained for 200 epochs, the 4mer model for 100 epochs, and the 5mer model for 25 epochs. These epoch numbers were chosen because they hit a sweet spot where the models achieved satisfactory F1 scores without excessively prolonging the training time. To ensure consistency in the comparison, each 3mer model was trained for the same number of epochs, and the same approach was followed for the 4mer and 5mer models. While it's possible that some models might require more or fewer epochs for optimal performance, this consideration was not explored as a standardized epoch count was used for each model.