mjpsm / Confidence-XGB

mjpsm/Confidence-XGB

What it does
Predicts a person’s confidence level — Low (0), Medium (1), or High (2) — from 10 numeric/ratio features.

Input Features (and how ratings affect predictions)

1. self_efficacy (1–10) — belief in one’s ability. Higher → more likely High .
2. past_success_rate (0–100) — % of goals achieved. Higher success → higher confidence.
3. preparation_time_hours (0–40) — average prep hours. More prep → higher confidence (with plateau).
4. feedback_positive_ratio (0–1) — positive / total feedback. Higher → higher confidence.
5. support_system_strength (1–10) — perceived support. Higher → higher confidence.
6. stress_management (1–10) — coping under pressure. Higher → higher confidence.
7. growth_mindset (1–10) — belief skills improve with effort. Higher → higher confidence.
8. resilience_score (1–10) — bounce-back ability. Higher → sustains confidence after setbacks.
9. task_familiarity (1–10) — experience with task type. Higher → higher confidence.
10. task_difficulty_perception (1–10) — perceived difficulty. Higher → lower confidence.

Labels

• 0 → Low, 1 → Medium, 2 → High
  (See label_map.json .)

Training Summary

• Framework: XGBoost (hist)
• Data: 3,000 rows (balanced 1,000/class)
• Split: 70/15/15, early stopping=100
• Metrics (test): Acc 0.982 , Macro-F1 0.982 , LogLoss 0.046
  See training_summary.json ; confusion matrix in confusion_matrix.png .

Load & Predict

from huggingface_hub import hf_hub_download
from xgboost import XGBClassifier
import json, pandas as pd

REPO = "mjpsm/Confidence-XGB"

model_file = hf_hub_download(REPO, "xgb_model.json")
feat_file  = hf_hub_download(REPO, "feature_order.json")
map_file   = hf_hub_download(REPO, "label_map.json")

clf = XGBClassifier(); clf.load_model(model_file)
features = json.load(open(feat_file))
label_map = json.load(open(map_file))
inv = {v:k for k,v in label_map.items()}

x = pd.DataFrame([{
  "self_efficacy": 8, "past_success_rate": 76, "preparation_time_hours": 12,
  "feedback_positive_ratio": 0.72, "support_system_strength": 7, "stress_management": 6,
  "growth_mindset": 9, "resilience_score": 8, "task_familiarity": 7, "task_difficulty_perception": 4
}])[features]

p = clf.predict_proba(x); pred_id = int(p.argmax(axis=1)[0])
print("Pred:", inv[pred_id], "(id:", pred_id, ")", "proba:", p[0].round(4).tolist())