Projects

AI forecasting tool based on R Shiny architecture, where you can upload your data, edit your data & visualize, select desired model, predict & forecast and finally extract. Also, it let you discuss on multiple topics with AI. 

Analysis of WPI Time Series Data

Here we have performed detailed study, modelling & forecasting of WPI (Wholesale Price Index) time series data. The data was available from April 1994 to January 2020, but there the main difficulty is the various base year i.e. the data was available from April 1994 to August 2010 with base year 1993-94, similarly from January 2005 to December 2012 with base year 2004-2005, again from April 2012 to January 2020 with base year 2012. So, to perform analysis approximated the whole data to the base year 2012 by backward regression approximation. Now here we don’t know whether the data follows additive model or multiplicative model, so we will consider both data and its logarithm where the logarithm of data will play the role for multiplicative model. Next tried to fit 3 different models and based on minimum AIC (Akaike’s Information Criteria) choosed the best fitted model and using that model forecasted the WPI for February to October 2020 for both data. Next, we studied for various time series outliers like innovational outliers, additive outliers, level shift outliers, transient outliers, seasonal level shift outliers etc. Next, we studied structural breaks and fitted a model on last break points to end of data and forecasted from February to October 2020 and tried to compare it with the model forecasting.

ANALYSIS OF INDEX OF INDUSTRIAL PRODUCTION OF INDIA (IIP)

Here the data for Index of Industrial Production of India (IIP) available from April 1993 to March 2019. Since, here the data was available for different base periods, we transformed the whole data to the same base period. Next, we had studied the behavior of IIP (e.g. trend, seasonal fluctuations, pattern in stationary part etc.) and we tried to fit 4 different models to both original series & logarithmic series and choose the best models for each series based on minimum Akaike’s Information Criterion (AIC). Next we forecasted based on these best fitted models and studied further for presence of structural breaks and outliers.