Coconut Price Prediction in Sri Lanka Using Supervised Machine Learning Approach (LSTM)

Abstract

Among Sri Lankans Coconut is a very prominent food with versatile usages. Coconut price is a significant factor but unfortunately the coconut price is fluctuating unpredictably because of several factors like rainfall and Soil Conditions. The goal of this project is to create a machine-learning model that uses supervised learning to estimate the price of coconut in Sri Lanka. This model can be used by the Coconut Manufacturing sector and Sri Lankan farmers to forecast prices and take required production actions. Coconut Development Authority information was selected for this research. The review of the literature assisted in identifying past research in relation to previously built crop price prediction models. The weekly coconut price data set was used to create and test the supervised learning model, and it is accessible from the Coconut Development Authority Web Site. For the modeling approach, a cross-industry data mining standard method was applied. It entails the processes of business comprehension, data comprehension, data preparation, modeling, and development. By analyzing the dataset and literature reviews it was decided to use the univariate Time series. long-short term memory neural network (LSTM) was selected as best fit for the scenario because it has eliminated the long-term dependency problem. Long Short-term memory has some variants named as Vanilla LSTM, Stacked LSTM and Bi-directional LSTM. As a first step the data was checked for the null values and prepared the dataset for model creation. The dataset was divided to two sectors as train and test data and Variants LSTM models are created and tested for different parameters. 50 to 250 epochs were changed during the model training and recorded the results. These results were evaluated using mean squared error and mean absolute error. According to the results that evaluated by mean squared error and mean absolute error measures shows that the Bi-directional LSTM was shows the best results. Finally using the results, it was concluded that for the Coconut price prediction Bi-directional LSTM is ideal. finally, for access the results to the users the power bi tool was used and published the results through web portal.