1Department of Agricultural Statistics, Bidhan Chandra Krishi Viswavidyalaya, Mohanpur-741 252, West Bengal, India
2Centre for Management of Health Services, Indian Institute of Management, Ahmedabad-380 015, Gujarat, India
3Department of Data Science, Prasanna School of Public Health, Manipal Academy of Higher Education, Manipal-576 104, Karnataka, India
Dairying in India has witnessed a radical transformation from a largely unorganised activity into a thriving organised industry. However, there are only a limited number of earlier attempts that specifically evaluated the milk production pattern of India. Moreover, most of these earlier studies are quite dated and have employed autoregressive integrated moving average (ARIMA) models, which suffer from the problem of local optima. To overcome this lacuna, we have utilised a genetic algorithm (GA) in the ARIMA framework. Its suitability to forecast the annual milk production in India has also been assessed comparatively with respect to the traditional ARIMA methodology.
For the current study, data on annual milk production (in million tonnes) in India for the period from 1980 to 2019 have been utilised. For both the approaches under investigation, the whole data series is first divided into two sets, namely the training set and the testing set. The production data of 1980-2016 have been utilised for the model building purpose while retaining the last 3 years’ data for the post-sample evaluation.
Outcomes emanated from the post-sample assessment clearly suggest that the ARIMA-GA approach has outperformed the traditional ARIMA methodology in terms of root mean square error (RMSE) and mean absolute percentage error (MAPE) values. It is also evident that GA has substantially minimised the error related to the parameter estimation.
ARIMA, Evolutionary algorithms, Forecasting, GA, Milk production
