akhyar919/DeBERTa-V3-Base-Phishing-Email-Classification

Model Card for Model ID

This model is finetuned version of DeBERTa-V3-base on a dataset from Kaggle. It is a text classification model which classifies phishing and safe emails.

Model Details

Model Description

• Developed by: Akhyar Ahmad
• Model type: Text Classification Model
• Finetuned from model: DeBERTa-V3-base

Uses

It is a text classification model which classifies phishing and safe emails. It can be used to seperate emails in bulk.

Training Details

Training Data

Training Data was from Kaggle having aroung 18k+ samples. which were shorted to 15k

Training Procedure

Training notebook is availabel on my GitHub repo: https://github.com/ico-akhyar/LLM-Finetuning/tree/main/Phishing_Email_Detection

Training Hyperparameters

learning_rate : 2e-4 per_device_train_batch_size : 8 gradient_accumulation_steps : 4 num_train_epochs : 3 fp16 : True

Lora Params

r : 16 lora_alpha : 32 lora_dropout : 0.05

Training

Epoch: 1 Training_Loss: No log Validation_Loss: 0.072302 Accuracy: 0.9703333333333334 F1: 0.9704321825452422 Recall: 0.9703333333333334

Epoch: 2 Training_Loss: 0.163300 Validation_Loss: 0.163300 Accuracy: 0.9746666666666667 F1: 0.9747319948425125 Recall: 0.9746666666666667

Epoch: 3 Training_Loss: 0.051300 Validation_Loss: 0.051300 Accuracy: 0.974 F1: 0.9740544607033279 Recall: 0.974

Results

Accuracy: 97.4667%

F1: 97.4732%

Recall: 97.4667%

Environmental Impact

• Hardware Type: Nvidia T4 GPU
• Hours used: <1h
• Cloud Provider: Google Colab

Framework versions

• PEFT 0.17.1