ARIMA-Analysis

This project uses Python to perform a time series forecasting using the ARIMA (AutoRegressive Integrated Moving Average) model on multiple datasets for a client that wanted to see if they could forecast the number of final registrations for a conference based on the way the registrations are being made prior to the stat date of the conference.

There are 2 sets of code:

• ARIMA Parameters
• ARIMA Code

The ARIMA Parameter code determines the best ARIMA parameters for each dataset and prints out the results. The ARIMA Code then implements the best parameters to forecast the data.

ARIMA PARAMETERS

Importing Libraries:

• The code imports pandas for data manipulation, pmdarima for automatic ARIMA parameter selection, and matplotlib for plotting ACF and PACF plots.

determine_arima_parameters Function:

• This function takes a file path as input.
• It loads the dataset, converts the 'Created Date' column to datetime format, and aggregates the data to obtain daily registrations.
• ACF and PACF plots are generated for manual inspection, providing insights into the autocorrelation and partial autocorrelation structures in the data.
• The auto_arima function from pmdarima is then used to automatically determine the best parameters for the ARIMA model.
• The determined ARIMA parameters are returned.

ARIMA Parameter Determination:

• The code iterates through each dataset file path specified in file_paths.
• For each dataset, it calls the determine_arima_parameters function to determine the best ARIMA parameters.
• The determined parameters are printed, indicating the optimal configuration for the ARIMA model for each dataset.

ACF and PACF plots:

• The ACF (AutoCorrelation Function) plot shows the correlation between the time series and its lagged values at different lags.
• It helps identify potential autoregressive (AR) terms in the ARIMA model by indicating the significant lags where the correlation is strong.
• The PACF (Partial AutoCorrelation Function) plot displays the correlation between the time series and its lagged values after removing the effects of intervening lags.
• It assists in identifying potential moving average (MA) terms in the ARIMA model by highlighting significant lags where the correlation is strong, excluding the influence of other lags.

ARIMA CODE

Importing Libraries:

• The code imports necessary libraries such as pandas for data manipulation, numpy for numerical computations, statsmodels for time series analysis, matplotlib for plotting, and sklearn for error metrics computation.

forecast_with_arima_adjusted Function:

• This function takes two arguments: file_path (path to the dataset file) and params (dictionary containing ARIMA model parameters for different datasets).
• It loads the dataset from the specified file path, converts the 'Created Date' column to datetime format, and groups the data by date to get daily registrations.
• Ensures that the time series is continuous with no missing dates by reindexing.
• Splits the data into training and testing sets using an 80% threshold.
• Fits an ARIMA model to the training set using parameters specified in the params dictionary.
• Forecasts for the length of the test set using the fitted model.
• Performs Augmented Dickey-Fuller (ADF) test for stationarity on the training data.
• Calculates error metrics such as RMSE, MAE, and MAPE.
• Computes AIC and BIC values.
• Plots historical data, actual test data, and forecasts.
• Returns the forecast.

file_paths and arima_params:

• file_paths contains the file paths for the datasets.
• arima_params is a dictionary containing the best ARIMA model parameters for each dataset.

Forecasting for Each Dataset:

• It iterates through each dataset file path.
• Extracts the dataset name from the file path.
• Prints the dataset name.
• Calls the forecast_with_arima_adjusted function with the dataset file path and corresponding ARIMA parameters.
• Stores the forecast results in a dictionary.

Output:

• The code prints forecast results, including ADF statistic, p-value, error metrics (RMSE, MAE, MAPE), AIC, and BIC for each dataset.
• It plots the historical data, actual test data, and forecasts for each dataset.