panchsan123/calories-burned-predictor

🔥 Calories Burned Predictor

A lightweight PyTorch regression model that predicts calories burned during a workout from basic biometrics and exercise stats.
Built as part of the learn-pytorch-2026 project.

Live demo → 🤗 panchsan123/calories-burned-demo

---

Model Description

CalorieModel is a fully-connected feedforward neural network trained for regression on the Kaggle Calories Burned dataset.
It takes 7 input features and outputs a single continuous value (calories burned in kcal).

---

Architecture

Input(7)
  → Linear(128) → ReLU → Dropout(0.2)
  → Linear(64)  → ReLU → Dropout(0.2)
  → Linear(32)  → ReLU → Dropout(0.2)
  → Linear(1)                           ← no activation (raw regression output)

Property	Value
Total parameters	~10,625
Loss function	MSELoss
Optimizer	Adam (lr = 0.001)
Scheduler	ReduceLROnPlateau (patience=10, factor=0.5)
Early stopping	patience = 15 epochs

---

Input Features

Inputs must be standardized using the scaler_mean and scaler_std arrays stored in config.json before inference.

#	Feature	Type	Description
0	Gender	int	1 = Male, 0 = Female
1	Age	float	Years
2	Height	float	Centimetres
3	Weight	float	Kilograms
4	Duration	float	Workout duration in minutes
5	Heart_Rate	float	Average BPM during workout
6	Body_Temp	float	Body temperature in °C

---

Training Details

Setting	Value
Dataset	Kaggle — "Calories Burned during Exercise"
Dataset size	15,000 rows, 0 missing values
Split	70% train / 15% val / 15% test
Preprocessing	StandardScaler fit on train only
Max epochs	150
Actual stop	Epoch 70 (early stopping)
Batch size	64
LR schedule	0.001 → 0.0005 (epoch ~30) → 0.00025 (epoch ~55)
Hardware	Google Colab T4 GPU
Seed	42

---

Evaluation Results (Test Set)

Metric	Value	Meaning
RMSE	1.79 kcal	Average prediction error within 1.79 calories
MAE	1.23 kcal	Typical prediction off by 1.23 calories
R²	0.9992	Model explains 99.92% of variance in calorie burn

---

Files in this Repo

File	Format	Use
best_model.pt	PyTorch state_dict	Reload and fine-tune in Python
model_scripted.pt	TorchScript	Android / Java via PyTorch Mobile
model.onnx	ONNX	Maximum portability — C++, Java, web, edge
config.json	JSON	Scaler params, feature order, metrics

---

How to Use

Python (TorchScript)

import torch, json, numpy as np
from huggingface_hub import hf_hub_download

# Download model and config
config_path = hf_hub_download("panchsan123/calories-burned-predictor", "config.json")
model_path  = hf_hub_download("panchsan123/calories-burned-predictor", "model_scripted.pt")

with open(config_path) as f:
    cfg = json.load(f)

model = torch.jit.load(model_path, map_location="cpu")
model.eval()

def predict_calories(gender, age, height, weight, duration, heart_rate, body_temp):
    gender_encoded = 1 if gender.lower() == "male" else 0
    raw    = np.array([[gender_encoded, age, height, weight,
                        duration, heart_rate, body_temp]], dtype=np.float32)
    mean   = np.array(cfg["scaler_mean"])
    std    = np.array(cfg["scaler_std"])
    scaled = (raw - mean) / std

    with torch.no_grad():
        result = model(torch.FloatTensor(scaled)).item()
    return round(result, 1)

# Example
print(predict_calories("Male", 28, 175, 72, 30, 145, 40.5))
# → ~260.4 kcal

config.json structure

{
  "features":     ["Gender", "Age", "Height", "Weight", "Duration", "Heart_Rate", "Body_Temp"],
  "scaler_mean":  [...],
  "scaler_std":   [...],
  "gender_map":   {"male": 1, "female": 0},
  "architecture": {"hidden": [128, 64, 32], "dropout": 0.2},
  "metrics":      {"rmse": 1.79, "mae": 1.23, "r2": 0.9992}
}

---

Limitations

• Trained on a single Kaggle dataset — predictions may be less accurate for athletes or people with unusual physiology
• No activity type input (running vs cycling vs weightlifting are all treated the same)
• Body temperature during exercise is not commonly measured — the live demo uses a pre-filled typical value

---

Citation

@misc{calories-burned-predictor-2026,
  author  = {panchsan123},
  title   = {Calories Burned Predictor},
  year    = {2026},
  url     = {https://huggingface.co/panchsan123/calories-burned-predictor}
}