Abstract:
Sustainable agriculture is a necessity to overcome global and local food security
challenges. At the same time, productive agriculture is also important to enhance the
socio-economic status of farmers. Integrating modern technologies with the
agricultural sector was identified as one of the most important solutions to overcome
many issues. Therefore, we aimed to apply machine learning technologies to identify
the most suitable crop types for productive farming. Badulla district was the focused
area for these studies. Potato, tomato, green gram and red onion were the selected
crop types. Rainfall, minimum and maximum temperature, minimum and maximum
relative humidity were the selected weather conditions. Wholesale price and retail
price of each of the above crop type were considered crop prices. Locations were
specified as gramasewa divisions and their soil types were considered. CRISP-DM
methodology was followed throughout the research. Weka libraries were integrated
with Java programming language for the implementation and MYSQL database was
used with JDBC database connector to maintain the data. Data mining classification
technologies were trained and tested in different conditions while performances were
evaluated using mean absolute error values and root mean squared error values. M5P
model tree and Random forest tree performed comparatively better performances in
weather forecasting and crop prices forecasting. In the system, farmers have to select
the relevant gramasewa division for their farming location. The system will identify
specific soil types in the relevant land and identify suitable crop types. Then,
predicted weather conditions are compared with required weather conditions for
each crop. Finally, the crop prices were evaluated. According to the results, the
higher-ranking crop list was provided to farmers as the output. These results may
help in decision making in the crop selection process while contributing to change
the field of agriculture as a profitable industry.
