Avocado Price Forecasting — Time Series Analysis

Forecasts weekly conventional avocado prices across the US (2015–2018) using SARIMA. Structured as a modular Python pipeline, not a notebook — each stage is a separate module with a single entry point.

---

Model Performance

Metric	Score	Interpretation
RMSE	0.22	±$0.22 average error
MAE	0.18	Median absolute deviation
MAPE	13.58%	~86.5% directional accuracy

---

Project Structure

Time-series-analysis/
├── data/
│   └── avocado.csv          # Raw dataset (Kaggle — not committed)
├── src/
│   ├── main.py              # Entry point — runs full pipeline
│   ├── loader.py            # Data ingestion and preprocessing
│   ├── model.py             # ADF stationarity test + SARIMA training
│   ├── plots.py             # EDA and forecast visualisations
│   └── pics/                # Auto-generated charts (git-ignored)
├── notebooks/
│   └── exploration.ipynb    # EDA sandbox
├── requirements.txt
└── README.md

---

Setup

git clone https://github.com/KonulJ/Time-series-analysis.git
cd Time-series-analysis
pip install -r requirements.txt

Download avocado.csv from Kaggle and place it in data/.

---

Run

python src/main.py

The pipeline runs four stages in sequence:

1. Load and preprocess — filter TotalUS conventional, resample to weekly, interpolate gaps
2. EDA — generate 5 charts (trend, rolling stats, decomposition, distribution, ACF/PACF)
3. Stationarity — ADF test to confirm differencing requirement
4. SARIMA — 80/20 train/test split, fit (1,1,1)x(1,1,1,52), evaluate, save forecast chart

All charts are saved to src/pics/.

---

Key Concepts Demonstrated

Concept	Implementation
Time series preprocessing	Resampling, interpolation, stationarity testing (ADF)
Seasonal decomposition	statsmodels.tsa.seasonal.seasonal_decompose — trend + seasonality + residual
SARIMA modelling	SARIMAX(1,1,1)(1,1,1,52) — weekly seasonality captured
Forecast evaluation	RMSE, MAE, MAPE with 95% confidence intervals
Modular pipeline	No notebooks in production path — clean src/ with single entry point

---

Roadmap

• Prophet and LSTM comparison
• Auto hyperparameter tuning (auto-arima)
• Streamlit dashboard for interactive forecasting

---

Data

Avocado Prices — Kaggle | Hass Avocado Board, 2018

---

Built by Konul Jafarova