This research employs a layered hidden markov Model (LHMM) featuring Markovian emission states to investigate the intricate relationships among dietary practices, insulin administration, and glucose levels in individuals with diabetes. The forward algorithm is utilized to compute the likelihood of a sequence of emission states, while the Baum-Welch algorithm is implemented for parameter estimation. The analysis identified the probability mass function for abnormal and normal emission states across two three-state sequences. The findings indicate a considerable probability of commencing with an imbalanced nutritional status, underscoring the need for dietary modifications. Transition probability matrices illustrate the dynamic interactions between food intake and insulin usage, with emission probabilities highlighting the critical role of insulin therapy in sustaining normal glucose levels. The probability mass functions reveal a concerning trend of increasing abnormal states over time. Compared to the two-layer HMM, this model shows a high probability for the normal state. Simulation outcomes validate the model’s applicability, suggesting its potential use in clinical environments for monitoring diabetes management. The study offers valuable insights into diabetes care and emphasizes the necessity for ongoing monitoring and dietary adjustments.
Diabetes is a metabolic disorder that can be regulated through proper physical activity and diet management. This paper is about developing a stochastic model by using Layered Hidden Markov Model (LHMM) for diabetes. In LHMM, we considered two hidden layers and one emission state layer. Here, we find out the probability of emission state sequence by the forward algorithm. Further, we derived probability mass function for one-state, two-state and three-state sequences for non-diabetes and diabetes emission states. We also derived the explicit relations for mean, variance and other Pearson’s coefficients of Skewness and Kurtosis of the formulated probability distribution. The model behaviour is studied in detail with real-time data on diabetes from secondary data sources. This study considered three variables, namely (i) physical activity, (ii) obesity category based on BMI, and (iii) having diabetes. It is assumed that diabetes is influenced by obesity, and obesity is influenced by physical activity. After a thorough analysis of data, it is observed that (i) from one day sequence, the state of non-diabetes is more likely, (ii) from the two-day sequence, the chances of the probability for two consecutive occurrences of the emission state of non-diabetes is 0.50; and (iii) from the three-day sequence, the chance of three consecutive occurrences of the emission state non-diabetes is around 0.38. From the transition and emission probability matrices, we found that those who are doing physical activity daily have less chance of obesity, and those who are not doing physical activity have a high chance of obesity, and later, they will be getting the risk of diabetes
8 October 2024 in International Journal of System Assurance Engineering and Management