Anfis Modelling On Diabetic Ketoacidosis For Unrestricted Food Intake Conditions
Diabetic ketoacidosis is a complication of diabetes that occurs when body cannot produce insulin necessarily to convert glucose into energy, instead fat is used as energy source and produce ketone as a byproduct. Ketones can be detected in urine compounds, especially when there is a large number of...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/20757/1/Anfis%20Modelling%20On%20Diabetic%20Ketoacidosis%20For%20Unrestricted%20Food%20Intake%20Conditions%20-%20Galuh%20Wilujeng%20Saraswati%20-%2024%20Pages.pdf http://eprints.utem.edu.my/id/eprint/20757/2/Anfis%20Modelling%20On%20Diabetic%20Ketoacidosis%20For%20Unrestricted%20Food%20Intake%20Conditions.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Diabetic ketoacidosis is a complication of diabetes that occurs when body cannot produce insulin necessarily to convert glucose into energy, instead fat is used as energy source and produce ketone as a byproduct. Ketones can be detected in urine compounds, especially when there is a large number of ketones that produce a distinctive smell of acetone. Odor sensors assembled into Electrical Nose (E-nose) system is used as self-diagnostics pre-test for diabetic’s analysis. However, diabetic’s analysis often required a subject to fast before sample testing. Currently, different prediction model for diabetic ketoacidosis are used depending on fasting or non-fasting conditions. This is inconvenience for diabetic’s analysis
to be done at any time anywhere. This project aim to propose an adaptive prediction model capable to diagnose diabetic ketoacidosis in unrestricted food intake conditions. The adaptive Neuro-fuzzy Inference System (ANFIS) is proposed to build the diabetic ketoacidosis classifier. The fuzzy inference model will be used to capture both fasting and non-fasting membership functions before feeding the results for classification to the neural
network model. Two sets of experimental data involving 20 diabetic patients and 20 healthy subjects were collected from CITO laboratory Semarang Central Java, Indonesia. Ethics consents were informed and agreed by the subjects before starting the data collection. This project follows the experimental methodology in verifying the hypothesis drawn. The experimental paradigm was designed to simulate fasting and non-fasting conditions. Samples data were recorded in the morning before food intake and two hours after food intake using four MQ 2, MQ 5, MQ 6 and MQ 8 sensors, in previously built Electronic Nose prototype system. A 5-fold cross-validation testing was implemented for classification results analysis. The results are highly promising with at least 90% accuracy in all testing. The proposed model has achieved 96% average accuracy in unrestricted food intake conditions. The prediction results on non-fasting and fasting data samples were recorded as
98% and 96% of average accuracy respectively. This has proven that the proposed ANFIS model is good to detect diabetic’s cases through ketoacidosis regardless of food intake. It has better performance in normal food intake as compare to fasting condition, since insulin inefficiency happened in diabetics patients will resulted in obvious acetone secretion in nonfasting condition. The project has also implemented the optimization process onto the proposed ANFIS model through the hybrid of Genetic Algorithm on the fuzzy membership function of the ANFIS model. The proposed GA-ANFIS approach provides excellent classification in accuracy, precision and recall. However, the results are only a minor improvement from the non-optimized ANFIS model since the predecessor has achieved good classification accuracy. In conclusion, diabetic ketoacidosis in unrestricted food intake conditions can be predicted using the proposed ANFIS and GA-ANFIS model. Future work should be focusing on data collection of the E-Nose sensors and the improvement of the learning algorithm robustness towards environmental noise during data acquisition, such as
evaporation and contamination of odor samples. |
|---|