For a multi-threaded application, I want to identify the nature of the application based on the arrival times of each thread. Example(Are thread launch spaced regularly, are they bursty in nature or are they highly dynamic). To identify this, the starting time of each thread is measured. The variance is calculated for time taken between every thread launch. This gives an idea of the spread in inter-arrival times. More spread indicates high dynamism. Based on the percentage of zscores that fall within the first level of standard deviation, a prediction is made on the level of dynamism (if more than 50% of values falls within one standard deviation - low dynamism otherwise high dynamism). Is this approach correct? Also, the values obtained for the starting time of each thread is only for a sample and not for the whole population. For this reason, the standard deviation is calculated using n-1. Are there any exisiting techniques for better estimation? Would clustering the values based on similarity help in any way? I'd highly appreciate it if you could point me to any reference. Thanks in advance.
If your prediction is specifically "low-dynamism" vs. "high-dynamism" you could run a logistic regression on your observations. This would give you effectively a percentage probability of one of the two outcomes you need to forecast. Logistic regression (or any kind of regression for that matter) includes an analysis of the IVs distribution and significance, similar to the z-score analysis you describe in your question.
But the analysis you're doing now sounds reasonable if you simply want to look at the dispersion of the threads start. The unbiased sd estimator n-1 is only required if you have a small number of observations, say < 50.