DEEP LEARNING TIME-SERIES CRUDE PALM OIL PRICE FORECASTING MODEL WITH COMMODITIES, WEATHER AND NEWS SENTIMENTS.

Kanchymalay, Kasturi (2020) DEEP LEARNING TIME-SERIES CRUDE PALM OIL PRICE FORECASTING MODEL WITH COMMODITIES, WEATHER AND NEWS SENTIMENTS. Doctoral thesis, Universiti Teknologi Malaysia.

Text kasturipcs143048d20ttt.pdf Restricted to Registered users only Download (5MB)

Abstract

Crude Palm Oil (CPO) is an important commodity for the Malaysian economy, and thus the ability to forecast the future price of CPO would have a significant impact on the economic growth of the country. However, variations observed in the prices of other commodities, the unpredictable climate changes and sentiments in news affect CPO forecasting model. Existing traditional prediction models failed to capture both long and short term temporal behaviour, and hence resulting in imprecise CPO price forecasting. In this study, the influencing factors of the CPO price movement are explored and collected using multimodal data acquisition techniques. A combination of factors are integrated with CPO price movements for forecasting purposes: other commodities’ price (soybean oil price, rapeseed oil price, corn oil price, olive oil price, coconut oil price and crude oil price); weather elements (maximum temperature, minimum temperature, humidity, rainfall and wind speed); and news sentiments. Preliminary analysis of this study indicated a strong correlation between the CPO price and other commodities’ prices, between CPO price and weather elements, as well as between CPO price and news headline sentiment. In the proposed approach, the Long Short Term Memory networks (LSTM) was utilised to form a stacked LSTM (SLSTM), in which it has two hidden layers whereby each layer contains multiple memory cells. The SLSTM was employed to forecast the CPO price by considering the impact of other commodities’ price, weather elements, and news headline sentiments. The results indicate that the SLSTM forecasting model with other commodities’ price produced significantly better results as compared to the SLSTM univariate CPO price forecasting model, SLSTM forecasting model with weather elements, and SLSTM forecasting model with news headline sentiment. The experimental results also show that the best forecasting performance is achieved when all factors were combined to predict the CPO price. The proposed SLSTM approach demonstrated better results in comparison to baseline approaches: feedforward neural network; deep neural network; and statistical models such as Autoregressive Integrated Moving Average (ARIMA), as well as Exponential Smoothing. The finding shows that the proposed SLSTM model has the potential to represent the actual scenario of CPO price trends due to the fact that it was able to produce results that are similar to the actual CPO price data. Therefore, the model can be effectively utilised by policymakers in the formulation of policies pertaining to palm oil demand and supply.

Item Type:	Thesis (Doctoral)
Subjects:	Technology > Engineering (General). Civil engineering (General)
Depositing User:	ENCIK SAIFUL FADZLY JAMALUDIN
Date Deposited:	07 Jul 2026 17:38
Last Modified:	07 Jul 2026 17:38
URI:	https://repositori.mohe.gov.my/id/eprint/316

Actions (login required)

View Item