Agriculture forms the backbone of Kenya's economy contributing 30% of the country's GDP. However, little is known about the prices, supply and demand of fruits and vegetables. Market price information is a very important agricultural concern which can be enhanced by big data and machine learning techniques like predictive analytics. This research work was based on developing an agricultural commodity price prediction application that uses a machine learning model on appropriate data to facilitate commodity price market information. The research investigated the applicability of advanced machine learning techniques through the study of different predictive analytics models for forecasting agricultural commodity prices with a specific focus on sukuma wiki in Nairobi, Mombasa and Kisumu Counties.
The forecasting approaches used were regression predictive modelling using linear regression, ensemble techniques using random forests algorithm and the popular gradient boosting algorithm and its variant, XGBoost algorithm. These algorithms were then compared to identify the best performing algorithm that gave the best Root Mean Squared Error (RMSE) accuracy and therefore the best sukuma wiki price prediction.
The research work utilized data-sets from multiple sources including the National Farmer Information Service (NAFIS) weekly crop information on sukuma wiki. Additionally, data from survey results collected from a CEDIA-Utawala value-chain study done in five Kenyan counties, weather conditions such as prevailing temperature and precipitation, and macro-economic data such as the consumer price index and prevailing inflation rates. The methodology involved the collection of relevant data sets, cleaning and preparing the data, training the models, model testing and improving the models with a view to selecting the best performing model.
The price prediction application was able to provide real-time insights and a visualization of the prices trends for sukuma wiki in the past and forecasts on future trends. This study will assist in driving commodity prices, demand and supply knowledge. It will lead to smart decision-making and promotes markets for farmers, consumers, processors, traders and policy makers.
Keywords: agriculture, machine learning, predictive analytics, sukuma wiki, prices, counties.
