metadata
language: en
license: mit
tags:
- text-classification
- finance
- banking
- bert
datasets:
- Dikshith4500/bank-transaction-categories-50k
SpendingCategoryBERT
Fine-tuned BERT for classifying raw bank transaction strings into 8 spending categories. Built as part of Foresight — a proactive AI financial OS with 12 LangGraph agents and 6 custom MCP servers.
Categories
| Label | Category | Example |
|---|---|---|
| 0 | grocery | WHOLEFDS MKT #1523 |
| 1 | restaurant | STARBUCKS #12345 |
| 2 | transport | UBER *TRIP |
| 3 | shopping | AMZN MKTP US*2K4J9 |
| 4 | entertainment | NETFLIX.COM |
| 5 | utilities | ATT*BILL PAYMENT |
| 6 | healthcare | CVS PHARMACY #1234 |
| 7 | general | VENMO PAYMENT |
Performance
| Metric | Score |
|---|---|
| Accuracy | 100% |
| F1 (weighted) | 1.0 |
| Precision | 1.0 |
| Recall | 1.0 |
Usage
from transformers import pipeline classifier = pipeline("text-classification", model="Dikshith4500/SpendingCategoryBERT") classifier("WHOLEFDS MKT #1523")
Training
- Base model: bert-base-uncased
- Dataset: Dikshith4500/bank-transaction-categories-50k
- Examples: 6,973 labeled bank transaction strings
- Epochs: 3
- Learning rate: 2e-5
- Hardware: T4 GPU