Student-driven machine learning models empower farmers

Vietnam is the second largest coffee exporter in the world and contributes more than half of the global Robusta supply. The coffee production in 2022/23 reached 29.75 million bags, of which Robusta accounted for more than 95 percent.

In the International Coffee Organisation's Annual Review 2021/2022, Vietnam was ranked first in coffee cultivation yield with 2.4 tons/hectare. The coffee yields in Vietnam are made up of Robusta, Arabica, Cherri, Moka, and Culi beans, which are among the most popular coffee beans farmed in Vietnam.

However, the prices of agricultural products, including coffee beans, are often unstable and can fluctuate sharply during times of bumper crops, which can cause significant impacts on farmers’ income and damage to the economy.

Aiming to address this issue, over the span of four months, a group of final year Bachelor of Information Technology students from the School of Science, Engineering & Technology, including Nguyen Hai Minh Trang, Doan Chanh Thong, Le Ngoc Nguyen Thuan, Nguyen Phuong Nam and Lam Tin Dieu, trained and evaluated six machine learning (ML) models to predict coffee prices, which may help Vietnamese farmers make informed decisions about their crops and plan accordingly, maximising their profits and minimising their losses.

“We developed six ML models, namely LSTM, GRU, ARIMA, SARIMA, SVM and RF, based on historical coffee prices, gasoline prices, temperature, and precipitation, to predict Robusta coffee prices in Lam Dong province and discovered that the RF model, which utilised the entire dataset, was the most effective,” said Trang. 

“It can incorporate richer dataset and handle nonlinear relationship.

“Additionally, fuel prices were found to be a significant predictor and outperformed across all other tested feature combinations.”

The team highlighted that the model has the potential to be continuously improved by studying and adding to the impact of crop yields, market trends and geopolitical events on the prices of agricultural products.

Each team member faced different challenges during their project, such as lack of deep understanding of various ML models, effectively conveying the complexities of their work to the AI domain, or time management and communication while being physically distant from their teammates. However, by investing considerable time in research, delving into AI and ML papers, and enhancing their technical skills and project coordination, they improved their AI research skills for real-world problems and potentially develop their research into practical products.

“Our primary challenge centred around data collection and integration,” Thuan said.

“While model development was straightforward, the substantial time invested in acquiring and incorporating data posed a significant hurdle.

“Each team member navigated a steep learning curve in both technical skills and project coordination. The process involved extensive research activities, fostering innovation and the creation of new solutions.”

Nam worked remotely from Hanoi and has had a full-time job while conducting the research. To prevent a potential slowdown and disruption in communication, Nam said the team established weekly meetings and maintained constant communication, motivating each other to stay on track and successfully tackle the workload.

The capstone project was closely supervised by the lecturers from the School of Science, Engineering & Technology, RMIT Vietnam. Its results were recently presented at a prestigious conference - The 8th IEEE/ACIS International Conference on Big Data, Cloud Computing, and Data Science Engineering (BCD 2023) - together with the established researchers, scientists, engineers, industry practitioners in the field of Big Data, Cloud Computing, and Data Science.

The team plans to refine their models based on feedback received during the presentation, and also aims to explore additional avenues for improving the accuracy and applicability of their predictions.

“We plan to dive into more advanced techniques and emerging methodologies within the field to enhance the robustness of our research outcomes,” Thong said.

“Collaboration with other experts in the domain and potential partnerships may also be pursued to expand the scope and impact of our findings.”

The team aims to continually iterate and advance their research to make meaningful contributions to the evolving landscape of Big Data and AI in the specific context of their research.

Story: Ha Hoang