Data Quarterly Series : 2 | Time Steps:33
Monthly Series : 3 | Time Steps:204
Tasks: Appropriate charts to understand the characteristics of the time series. Forecast of each of the 12 time steps for each series
Assumption: The solutions are based on the assumption that the two quarterly series and the three monthly series areunivariateinnature based on the problem statement. Exploratory Data Analysis
Modelling
Time Series Plotting:
• Purpose: Visual inspection of data to identify trends, seasonality, and irregular components. • Why: Initial step to understand data behavior over time.
Rolling Statistics:
•Purpose: To observe the moving average and moving variance, helping to spot trends and volatility changes. •Why: Helps in a preliminary check for stationarity.
Decomposition of Series:
•Purpose: Separation of data into trend, seasonal, and residual components. •Why: To understand underlying patterns and to aid in handling the stationarity in the data.
Stationarity Testing with ADF:
•Purpose: Applying Augmented Dickey-Fuller test to formally testfor the presence of a unit root.(non-stationarity) •Why: Confirming stationarity is crucial before applying certainstatistical models.
ACF and PACF Plots:
•Purpose: To identify the autocorrelation in the data which helps in selecting ARIMA model parameters. •Why: These plots determine the order of the AR and MA components in ARIMA models.
Parameter Selection with Auto ARIMA:
• Purpose: Automated search for the best ARIMA model parameters based on the data.
• Why: Optimizes model selection process by finding the best fit with minimal AIC.
SARIMA Model Predictions:
•Purpose: Applying Seasonal ARIMA models to forecast future data points with seasonality adjustments. •Why: SARIMA accounts for both seasonality and trends, providing more accurate forecasts.
Prophet
•Purpose: Prophet is compared with SARMIA method for quarterly data •Why: The stark decrease in the series values in Quartely series around 2020 calls for mapping these peculiar irregularities.