Maize crop yield prediction through reinforcement learning, artificial neural network and alert messages generation

Abstract

Most of the medium and small-scale Kenyan farmers rely on metrological department for weather information and subsequently also rely on the ministry of agriculture to provide information on the crop yield. This seldom occurs as the agricultural officials do not use the scanty available data for data mining to provide relevant information to farmers. While the current metrological stations are few, thus accurate weather and crop yield information for any season and station cannot be disseminated to the diverse population that needs the information. The purpose for this project is to address the temperature, rainfall and maize crop yield information to relevant bodies and to provide a centralized information dissemination center for farmers. To analyze, build, test and evaluate a prototype system that shall provide information to farmers and the meteorologists' officers. Agent based methodology was borrowed to develop the application prototype and thus the methodology used was Tropos methodology in combination to machine learning algorithm (Q learning). The implementation was carried out using Delphi 7 and MS SQL server 2008 and FANN Delphi components to exploit the artificial neural network features. Test cases were run for the purpose of evaluating the prototype in comparison to the actual data that was collected from the field. And thus graphical comparison of the test runs is as shown in the graphs and respective broadcast message generation and subsequent dissemination. The farmers and meteorologist too used the system