Creating a Predictive Model for Telecommunication Network Base Station Availability in Minna using Autoregressive Integrated Moving Average.

Creating a Predictive Model for Telecommunication Network Base Station Availability in Minna using Autoregressive Integrated Moving Average.

Abstract:

In the realm of telecommunication hardware and software, a standard of 99.999% (five ‘nines’) availability is upheld to ensure Mobile Network Operators (MNOs) provide high-level service delivery. However, MNOs in Nigeria and many sub-Saharan African countries struggle to meet the expected base station availability due to prolonged restoration times after outages.

This thesis examines the historical Base Transceiver Station (BTS) Availability reports of four MNOs (MNO W, MNO X, MNO Y, and MNO Z) in Minna, utilizing a dataset comprising a thousand data points for each MNO, gathered between January 1, 2018, and September 26, 2020. The dataset was divided into a Training period (73% of the data) and a Validation period (27%). The Time Series (TS) data was modeled using the Autoregressive Integrated Moving Average (ARIMA) prediction technique. ARIMA (p,d,q) parameters were determined through Correlation plots, Autocorrelation Function (ACF), and Partial Autocorrelation Function (PACF). The specific ARIMA models for MNOs were found to be ARIMA (0,1,3), ARIMA (1,0,1), ARIMA (2,0,4), and ARIMA (0,1,1) for MNO W, MNO X, MNO Y, and MNO Z, respectively.

The developed ARIMA-based predictive models were used to forecast BTS Availability for the MNOs from September 27, 2020, to December 20, 2020. To assess the model performance, Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) were employed during the Validation period. The calculated MAEs and MAPEs for the respective MNOs are as follows: 1.3959, 0.6602, 1.5666, and 0.6177; and 0.0150, 0.0068, 0.0176, and 0.0063.

Additionally, a Long Short-Term Memory (LSTM) model was applied for comparison with the ARIMA model on the same MNOs. The LSTM model yielded higher MAE and MAPE values, with differences of 51%, 26%, 44%, and 45% for MNO W, MNO X, MNO Y, and MNO Z, respectively. These findings indicate that the ARIMA models outperform the LSTM models for all the MNOs. The low MAE and MAPE values for the predictive models demonstrate their accuracy, making them suitable for informed decision-making and proper planning.

Moreover, the developed Predictive Maintenance (PdM) algorithm based on the 95% availability threshold allows MNOs to proactively schedule maintenance. While MNO W and MNO Y show no significant savings in maintenance count, MNO X and MNO Z experience savings of 33 and 32, respectively. These results provide valuable insights for MNOs to optimize their maintenance strategies and improve overall network performance.