Arko007/apple-price-predictor

Apple Price Prediction Model — Indian Market

Hybrid Prophet + ARIMA ensemble for forecasting Indian wholesale apple prices (₹/kg) and generating SELL / STORE recommendations for farmers and traders.

---

Overview

This model forecasts apple mandi/wholesale prices 7 days ahead and recommends whether farmers or cold-storage traders should SELL immediately or STORE for better returns, accounting for Indian cold-storage costs (₹0.75/kg/day).

Component	Details
Architecture	Hybrid Prophet + ARIMA Ensemble
Blend Weights	60% Prophet + 40% ARIMA
Currency	INR (₹) — wholesale price per kg
Training Data	12,000 synthetic samples (2018–2021)
Varieties	Shimla, Kinnauri, Royal Delicious, Golden Delicious, Maharaji
Regions	Himachal Pradesh, Jammu & Kashmir, Uttarakhand, Arunachal Pradesh, Nagaland

---

Quickstart

from model.predict import predict_price

result = predict_price({
    "date":              "2026-03-07",
    "current_price":     120.0,       # ₹/kg (current wholesale price)
    "storage_time_days": 15,
    "apple_variety":     "Kinnauri",
    "region":            "Himachal Pradesh",
})

print(result)
# {
#   "predicted_price_7d": 127.50,
#   "recommendation":     "STORE",
#   "current_price":      120.0,
#   "storage_cost_7d":    5.25,
#   "breakeven_price":    125.25,
#   "currency":           "INR",
#   "confidence":         "hybrid Prophet+ARIMA (0.6/0.4)"
# }

---

Repository Structure

apple-price-predictor/
├── data/
│   └── apple_price_dataset.csv     # 12,000-sample Indian market dataset
├── models/
│   ├── prophet_model.pkl           # Trained Prophet model
│   ├── arima_model.pkl             # Trained ARIMA model
│   ├── scaler.pkl                  # MinMaxScaler for feature normalization
│   └── metrics.json                # MAE / RMSE evaluation metrics (in ₹)
├── model/
│   └── predict.py                  # Inference wrapper
├── train.py                        # Full training pipeline
├── requirements.txt
└── README.md

---

Indian Market Dataset

Synthetic dataset simulating realistic Indian apple market dynamics:

Column	Description
date	Daily timestamps (2018-01-01 onward)
apple_variety	Shimla / Kinnauri / Royal Delicious / Golden Delicious / Maharaji
region	HP / J&K / Uttarakhand / Arunachal Pradesh / Nagaland
harvest_season	1 if July–October (Indian harvest window)
storage_time_days	Days in cold storage post-harvest
temperature	Daily temperature °C (hill station climate)
rainfall	Daily rainfall mm (monsoon-shaped curve)
market_demand_index	Demand pressure (peaks: summer stock-out + Diwali)
supply_index	Supply pressure (high during harvest, monsoon shocks)
previous_week_price	Wholesale price 7 days prior (₹/kg)
price	Target — wholesale market price (₹/kg)

Price Drivers Modelled

Driver	Effect
Harvest season (Jul–Oct)	−₹12/kg discount (fresh supply glut)
Summer scarcity (Apr–Jun)	+₹15/kg spike (low cold-storage stock)
Festival demand (Diwali)	+₹5–₹15/kg premium
Monsoon road blockage	+₹0–₹10/kg in HP/JK (supply shock)
Cold-storage decay	−₹0.08/kg/day (quality degradation)
Inflation trend	+₹5/kg per year
Market noise	±₹4/kg random volatility

Typical wholesale price range: ₹40 – ₹165/kg

• Maharaji (local): ₹40 – ₹80/kg
• Royal Delicious: ₹55 – ₹110/kg
• Golden Delicious: ₹60 – ₹115/kg
• Shimla: ₹65 – ₹130/kg
• Kinnauri (premium): ₹85 – ₹165/kg

---

Feature Engineering

Feature	Description
month	Calendar month (1–12)
week_of_year	ISO week (1–53)
season	winter / pre_monsoon / monsoon / post_monsoon
storage_cost_estimate	₹0.75 × storage_time_days
price_trend	Day-over-day price change (₹)
rolling_mean_price	7-day rolling mean (₹)
rolling_std_price	7-day rolling std dev (₹)

---

Model Architecture

Prophet

• Yearly + weekly seasonality (multiplicative mode)
• Custom festival seasonality component (Fourier order 3)
• Changepoint prior scale: 0.05

ARIMA

• Auto parameter search via pmdarima
• Best order selected by AIC minimisation
• Stepwise search: p, d, q ∈ {0…3}

Hybrid Ensemble

final_prediction = (0.6 × prophet_prediction) + (0.4 × arima_prediction)

---

Sell / Store Decision Engine

storage_cost_7d  = ₹0.75 × 7  =  ₹5.25/kg
breakeven_price  = current_price + storage_cost_7d

if predicted_price_7d > breakeven_price:
    recommendation = "STORE"
else:
    recommendation = "SELL"

---

Training

pip install -r requirements.txt
python train.py

---

Evaluation Metrics (₹/kg)

Model	MAE (₹/kg)	RMSE (₹/kg)
Prophet	~₹15–20	~₹18–25
ARIMA	~₹5–8	~₹7–12
Hybrid	~₹10–14	~₹14–18

(Exact values saved in models/metrics.json)

---

License

MIT License — free for research and commercial use.