Evaluations of oil palm fresh fruit bunches maturity degree using multiband spectrometer

The demand for high accuracy in grading system of the ripeness and maturity level of the Oil Palm Fresh Fruits Bunch (FFB) is one of the important impending factor facing the oil milling industry globally. This limitation directly affects the Oil Palm Extraction Rate (OER), the quantity, and quality of products produced annually, thereby reducing the profitability margin of the milling industry. However, numerous studies targeted at investigating several improved techniques of grading, and classification of the current oil palm maturity classification in order to enhance the oil palm yield in terms of quality, and quantity annually have been carried out. In view of the needed improvement in OER, there is need to test a novel data mining, and knowledge discovery approach within collected spectrometer data from the oil palm FFB, using WEKA software for classification, and prediction of the oil palm FFB ripeness, to enhance the current manual human grader system. In this research, several machine learning algorithms housed in WEKA data mining tool were proposed for the building of a classifier models, as compared with other earlier manual, and statistical analytical method which require high computational knowledge in coding, time consumption, and prion to human or computational error. This novel approach is tested on the 106-labelled sampled of oil palm FFB of four ripeness categories of unripe, under ripe, ripe, and over ripe by human grader, with reference to the stipulated standard of the Oil Palm Grading Manual (OPGM) of Malaysia. The reflectance data of different wavelength bands incidence from 8 LED modules upon the oil palm FFB, were extracted by 4 different sensors of a spectrometer in laboratory experiment, for onward detailed analysis with machine learning algorithms in WEKA data mining tool. The result illustrated that just one sensor features (sensor 4) are significantly enough to build a good, accurate classifier model that can predict, and classify the oil palm FFB ripeness, rather than the proposed 4 bands sensors with 32 feature attributes. Hence, reducing cost for other sensors, improving the analysis time for the classifier model building, and enhancing the productivity of the system at large. Furthermore, the Lazy-IBK algorithm have been validated to produce the best classifier model, with the machine learning algorithm performance of 65.26%, recall of 65.3%, and 65.4% F-measured as compared to other evaluated machine learning classifier algorithms proposed within the WEKA data mining algorithm. The ROC curve area indicated an average weighted value of 77.4% for the area under the curve as indicated, which is a measure applied for the accuracy of the applied algorithm. In conclusion, the simple machine learning algorithm model evaluation is developed to classify the oil palm maturity degrees, in order to validate the human grader assessments to enhance the productivity of the oil milling industries.

Preview Text FK 2017 66 - IR.pdf Download (1MB) | Preview

Actions (login required)

View Item