DNA_Sequence_Covid_Classification

This work is a sequence classification, we predict whether a DNA sequence belongs to the SARS-CoV-2 (Covid-19).

1. **Data description**
• Xtr.csv - the training sequences.
• Xte.csv - the test sequences.
• Ytr.csv - the sequence labels of the training sequences indicating Covid-19 DNA (1) or not (0).
• Xtr_vectors- the training vectors which contain the encoding vector of each DNA sequencing reads from training.
• Xte_vectors- the testing vectors which contain the encoding vector of each DNA sequencing reads from testing.
• Each row of Xtr.csv represents a sequence.
• Xte.csv contains 1000 test sequences, for which you need to predict the labels.
• Ytr.csv contains the labels corresponding to the training data, in the same format as a submission file.


2. **Pre-processing**
For features use the vectors dataset which are already encoded but still we need to convert the values 0 to -1. Therefore transforming {0,1} classification problem to {-1,1} classification problem. To do that we computed : $y=2y-1$


3. **Models**
We used many models to make classification like Logistic Ridge Regression (Newton-Ralphson ,SGD), Kernel Logistic Regression, Kernel Ridge Regression, Kernel SVM and HardMarginSVM. We got the best accuracy with using Kernel SVM using RBF kernel. Our model performs an accuracy of **100%** for training and **97%** for validation. We got this result by looking for the good hyper-parameters
• kernel = 'rbf'
• sigma =0.1407035175879397
• degree = 2
• C = 10.0
• tol = 1e-4