Optimal and efficient time series classification with burrows-wheeler transform and spectral window based transformation

Karthikeyan, T.; Sitamahalakshmi, T.

doi:10.21533/pen.v6i1.275

Cited by 1 publication

(1 citation statement)

References 22 publications

(28 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The GeneCycle R Package [19] developed for bioinformatics has general application for time series and was employed in this study. Some recent time series analysis research deals with data classification as well as data series use in environment, production and financial research [24], [25], [26]. In our paper we perform our own Vostok data analysis using R and Excel, in both time and frequency domains.…”

Section: Spectral Analysis Methodologymentioning

confidence: 99%

Time and frequency analysis of Vostok ice core climate data

Hodzic¹,

Kennedy²

2019

PEN

View full text Add to dashboard Cite

The periodicity of Vostok ice core climate temperature and gas concentration data indicate inherent long term past regularity of Earth's climate, with a period of around 100,000 years, warming around 15,000 and cooling of around 85,000 years. At this point we are at the top of one of the warming periods. Vostok data cover around 430,000 years, ie 4 climate cycles (warming-cooling), of similar but not quite the same duration. In this paper we perform a detailed time and frequency analysis of these data for each of the cycles as well as their various combinations, including a full tested period of 430,000 years. Time correlation analysis allows for more accurate time lag estimate in each cycle already noted between temperature change and carbon dioxide content. We estimate these lags to lie between 1000-2500 years, longer than previously concluded. On the frequency side we perform Fast Fourier Analysis and identify full spectrum of harmonics for various cycles, and then perform energy analysis to identify which of the harmonics contributes the most. The idea is to reduce the computational load for further modeling and analysis using Kalman Filter based prediction method. Once the prediction model is defined (a follow up paper) data will be split into two segments, Learning and Testing, in preparation of a Machine Learning fine tunning methodology. We can use last three or last two or even just last cycle to learn on, and then the current on going cycle to test on. This will result in real time prediction of relevant climate data. Assuming causal time regularity, more of these cycles are employed in training, more the prediction error for the next cycle should be reduced. Hence it is critical to perform very detailed time and frequency analysis of Vostok data as a solid data base for the prediction model to follow.

show abstract