Spam Detection Model

TF-IDF + Random Forest classifier for English email and SMS spam.

Model


Algorithm	Random Forest (100 trees, class_weight=balanced)
Features	TF-IDF, 3000 unigrams, min_df=2
Training rows	235,449
F1 (random split)	0.974
Language	English only

Files

file	description
`model_parts/part_aa` .. `part_ay`	the model, split into 25 parts
`confusion_matrix.png`	confusion matrix on the test split
`skeleton.py`	training script

The model is stored as 25 parts because a single-file upload of 232 MB was cut off mid-transfer. Each part is a byte-range chunk of spam_model.joblib.gz.

How to load

import gzip, joblib

parts = [f"model_parts/part_{s}" for s in
         ["aa","ab","ac","ad","ae","af","ag","ah","ai","aj",
          "ak","al","am","an","ao","ap","aq","ar","as","at",
          "au","av","aw","ax","ay"]]

blob = b"".join(
    hf_hub_download("Skibidibois2000/spam-detection-model", p)
    for p in parts
)
bundle = joblib.load(gzip.decompress(blob))

prob = bundle["model"].predict_proba(
    bundle["vectorizer"].transform(["your message"])
)[0][1]
print("SPAM" if prob >= bundle["threshold"] else "HAM")

With the huggingface_hub library:

from huggingface_hub import hf_hub_download

Note on reported accuracy

The 0.974 F1 comes from a random 80/20 split, which places near-duplicate messages from the same corpus on both sides. Treat it as an optimistic figure.

Downloads last month: -; Downloads are not tracked for this model. How to track