☕ Cafe Sales — Random Forest Regressor

A tuned Random Forest Regressor trained on cafe transaction data to predict Total Spent per transaction.

📊 Dataset

Property	Value
Source	Cafe Sales Dataset
Rows	9,540 transactions
Target	`Total Spent` (in currency units)

🧠 Features Used

Feature	Type
Item	Categorical (Coffee, Cake, Cookie, etc.)
Quantity	Numeric
Payment Method	Categorical (Cash, Credit Card, Digital Wallet)
Location	Categorical (In-store, Takeaway)
Day of Week	Numeric (0=Mon, 6=Sun)
Month	Numeric (1–12)
Is Weekend	Binary (0/1)

Price Per Unit was intentionally dropped to avoid trivial prediction (Total Spent = Qty × PPU).

⚙️ Best Hyperparameters (via RandomizedSearchCV, cv=5)

n_estimators      = 300
max_depth         = 10
min_samples_split = 10
min_samples_leaf  = 2
max_features      = 0.5

📈 Evaluation Metrics (Test Set — 20%)

Metric	Default RF	Tuned RF
MAE	0.7800	0.7007
RMSE	1.8726	1.6997
R²	0.8959	0.9142

🚀 Usage

import joblib
import pandas as pd
from huggingface_hub import hf_hub_download

# Model download karo
model_path = hf_hub_download(
    repo_id="rohansuyal/cafe-sales-rf-regressor",
    filename="rf_tuned_model.pkl"
)

# Load karo
model = joblib.load(model_path)

# Predict karo
sample = pd.DataFrame([{
    "Item": "Coffee",
    "Quantity": 2.0,
    "Payment Method": "Cash",
    "Location": "In-store",
    "Day_of_Week": 0,
    "Month": 6,
    "Is_Weekend": 0
}])

prediction = model.predict(sample)
print(f"Predicted Total Spent: {prediction[0]:.2f}")

🛠️ Tech Stack

Python 3.10
scikit-learn — RandomForestRegressor, Pipeline, ColumnTransformer
pandas / numpy — Data processing
joblib — Model serialization

📁 Files

File	Description
`rf_tuned_model.pkl`	Trained pipeline (preprocessor + model)
`README.md`	This file

Downloads last month: -