BERT-crime-analysis

Developer's note

Please do download and try out the model locally or on colab, as it helps huggingface determine that this model is important enough to have a serverless API for everyone to use. Also, the model is totally safe for everyone to use. The only reason one of the files has been marked unsafe because it is a pickle file.Thank you all for so much support !!

Overview

This model is a fine-tuned version of google-bert/bert-base-uncased on this Kaggle dataset. It achieves the following results on the evaluation set:

<!-- - Loss: 12.41% -->

• Macro f1: 89.44%
• Weighted f1: 93.15%
• Accuracy: 93.80%
• Balanced accuracy: 90.42%

Model description

This finetuned version of google-bert/bert-base-uncased excels at detecting the crime type from the description of the crime. It has 34 labels.

Training and evaluation data

<!-- - eval_loss: 12.41% -->

• eval_macro f1: 89.44%
• eval_weighted f1: 93.15%
• eval_accuracy: 93.79%
• eval_balanced accuracy: 90.42%

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30

Training results

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2
• Datasets 2.18.0
• Tokenizers 0.15.2