ZS-Data-Science-Challenge-2018

A national Data Science competition. (Ranked #18)

Method:

The statistical forecasting method used is the ARIMA time series with the regression model.

Step 1:

• Plot sales data as time series.

Step 2:

• Difference data to make data stationary on mean (remove trend). • This to remove the upward trend through 1st order differencing the series using the following formula.

Step 3:

• log transform data to make data stationary on variance.

Step 4:

• Difference log transform data to make data stationary on both mean and variance. • Formula: Y_{t}^{new'}=log_{10}(Y_t) -log_{10}(Y_{t-1})

Step 5:

• Plot ACF and PACF to identify potential AR and MA model. • Autocorrelation factor (ACF) and Partial autocorrelation factor (PACF) plots are used to identify patterns in the data obtained from step 4 which is stationary on both mean and variance. • The idea is to identify presence of AR and MA components in the residuals.

Step 6:

• Identification of best fit ARIMA model. • Auto arima function in forecast package in R helps us identify the best fit ARIMA model on the fly. • WE DO NOT NEED TO FEED THE DIFFERENCED DATA AS ARIMA AUTOMATICALLY DOES THAT. SO WE FEED ORIGINAL TIME SERIES AFTER STL() FUNCTION. • The best fit model is selected based on Akaike Information Criterion (AIC) , and Bayesian Information Criterion (BIC) values. • The idea is to choose a model with minimum AIC and BIC values.

Step 7:

• Forecast sales using the best fit ARIMA model. • Also I did, ACF and PACF for residuals of ARIMA model to ensure no more information is left for extraction.

Results Link: https://www.hackerearth.com/challenges/competitive/zs-data-science-challenge-2018/