Abstract:
Time Series data are of crucial importance as they depict trends among various entities over time. Finding similarity in patterns in time series data can help in getting meaningful insights in it. This work presents a technique that finds the similarity amongst time series data by taking distance between angular measure rather than absolute differences. Further, local patterns with relative increase or decrease are also included in computing over all distances. These distances can then be used in clustering or classification algorithms to find which data exhibit similar local patterns. The proposed technique aims in mining similarity between various time series by considering local patterns over global trend.
This technique was mainly developed for mining urban data, such as municipal budgets. City planning is mainly governed by the budget declared by the municipal corporation. Budget data can be utilized for finding how the city is progressing, what are the factors critical for its growth, why is it lagging in some factors as compared to other cities etc. To be able to derive meaningful conclusions, various mathematical and data mining techniques need to be applied on widely available municipal budget data. The budgets of Urban Local Bodies of India follow of uniform structure and the systematic analysis of them can give interesting insights. Although analyses of municipal budget data have been done in past but most of them have static nature and there is need for some generalized framework. This report presents techniques for broad analysis of Municipal Budgets by considering historical budget allocations as time series data.
The technique proposed is also applied on other time series data sets which comprises of both urban and non-urban (generic) data. The experiments gave better results as compared to the existing techniques
