The objectives of this research are to propose a proper model for risk assessment for drought in Thailand. A hidden Markov model is adopted. The results indicate that the amount of rainfall depends on the hidden states drought or non-drought state. If the state changes from drought to non-drought the amount of rain will increase. The probability of changing from drought state to non-drought state is higher than from drought state to drought state. The probability of drought occurrence in February is highest followed by in November March December January July April June October September August and May respectively. When the temperature increases the probability of drought occurrence will increase.

The objectives of this research are to propose a proper model for detecting defective machines in auto parts factories to apply the proposed model to the defective product data and to investigate the factors influencing on the machine producing defective products. A Bayesian hidden Markov model (HMM) is adopted. The results show that if the number of defective products increases the machine will be in the hidden bad state. The probability of changing from a hidden bad state to a hidden good state is higher than the probability of changing from a hidden good state to a hidden good state. The highest probability that the machine will be in a good state is at the work process 4 followed by 1 3 and 2 respectively. The highest probability that the machine will be in hidden good state is at the Worker no. 10 followed 9 3 12 7 2 1 6 11 4 8 and 5 respectively and The highest probability that the machine will be in the hidden good state is at the product type 1 followed by 2 and 3 respectively.

This research proposes a hidden Markov model for detecting machines with multiple fillers producing soft drink for quality control in order to reduce defective products. The proposed model is compared with a classical generalized estimating equations model (GEE). The monthly data the number of defective products the correctness and the time spending in correction were collected from a machine in a sample factory in Pratumthanee province. The two hidden states of the fillers are workable and unworkable. The states has a logistic distribution. The results show that the correctness of the filler error occurrence does not improve the probability of being in a unworkable state neither the period of time spending on correction. The current state seems to be the same as the previous state