forkedrepositories / indian-trade-data-analysis-and-forecasting

Indian-Trade-Data-Analysis-and-Forecasting

• Trade is an economic concept which involves Buying and Selling of the commodities, or exchanging goods and services between needy people.
• Trade is important in a way that it increases competition and decreases overall world wise cost of a product.

Data

• We scraped the trade dataset from Department of Commerce, Govt. of India website.
• Monthly Trade data is available from January, 2006 to January, 2020.
• We have total trade amount (Import/Export) for each month which is expressed in million US dollars.
• HS Code - Harmonized System (HS) of tariff nomenclature is an internationally standardized system of names and numbers to classify traded products, e.g. 1 for Live Animal, 95 for Toys, Games and Sports Requisites …
• Data can be viewed in terms of Country wise or HS code wise.

Model Used For Forecasting

• Exponential Smoothing
• Auto Regressive Model
• Moving Average Model
• Holt-Winters Model
• ARIMA Additive
• ARIMA Multiplicative
• ARIMA Seasonal
• RNN --> LSTM (Long Short Term Memory)

Forecasting values based on last 12 months trade amount, we predicted trade amount for last 13 months ie from Jan, 2019 till Jan, 2020.

Model Architecture For LSTM

Both models have same architecture : LSTM(200) --> LSTM(200) --> LSTM(150) --> DENSE(1)

Result

• Surprisingly Seasonal ARIMA outperformed LSTM in Forecasting Export Data, which shows that ARIMA captures seasonality much better than any other model, even with lesser data and randomness.
• LSTM model performed better for Import Data as expected because of it’s capability of retaining information of long period of time.