ICTworks

The Department of Agriculture Extension (DOAE) at the Ministry of Agriculture in Thailand is among the first government agencies in the country to adopt Artificial Intelligence technologies to real world applications. DOAE has developed such a system, namely, Personalized Data system, to deliver suggestions to millions of Thai farmers in a convenient, affordable and timely fashion.

Agricultural Extension Services via Artificial Intelligence

The objective of the Predicting Key Factors in Agricultural Databases for Thai Farmers to Make the Right Decision project was to bring advances in AI to provide personalized information to help millions of farmers in the country make better decisions, including what crops to grow, where to sell, etc. The idea was to present simple, precise and succinct enough information to the farmers. We needed to find an appropriate tradeoff between time required for computing and available computation time. The chosen algorithms must robustly provide good enough results in real world environment.

There are fifteen plants that enlisted as Thailand prime crops and the DOAE has seven large databases, collecting data from real world environment over several decades. We chose to present only rice’s because it is true main crop of the nation. While typical AI research expects experimental datasets and consumes a lot of time, this research investigated how to deliver the required agricultural extension capabilities with minimized training time. Three key factors, price, cost and yields of crops were analyzed, in order to make predictions simple but meaningful to farmers.

Very Accurate Agricultural Extension Services

We found that an artificial neural network with Multilayer Perceptron (MLP) and Random Forest (RF) models could effectively predict yield, cost, and price of crops. Adjusting parameters such as learning rate and the number of hidden nodes affect the accuracy of crop yield predictions. Smaller data sets required fewer hidden layers in model optimization. MLP models consistently produced more accurate yield predictions than RF models. Although considered a less accurate method, compared to MLP, RF works just fine in most cases.

MAPE (Mean Absolute Percentage Error) is used to measure both RF and MLP. It found that MLP models produced accurate predictions. Similarly, RF is used to provide suggestions when time constraints are tight. We found that the accuracy is very high, and MAPE is around 5% in most cases. In some difficult scenarios where data is not adequate, the results are still good, and MAPE is around 20%.

In the future, more advanced learning techniques can be analyzed and installed in the system. The involved factors should also be taken into account because there will also be more external data sources connected to the system.

A lightly edited synopsis of Predicting Key Factors in Agricultural Databases for Thai Farmers to Make the Right Decision by Chattrakul Sombattheera, Rapeeporn Chamchong and Phatthanaphong Chomphuwiset