Creating an Autoregressive Integrated Moving Average-Driven Predictive Model for Anticipating Base Station Availability



The telecommunications industry maintains a stringent standard of 99.999% (five nines) availability for both hardware and software systems to ensure optimal service delivery for Mobile Network Operators (MNOs). Nevertheless, MNOs in Nigeria and numerous sub-Saharan African countries struggle to meet the expected base station availability due to extended restoration times following outages. This research leverages historical Base Transceiver Station (BTS) Availability data encompassing one thousand data points for four MNOs: MNO W, MNO X, MNO Y, and MNO Z in Minna. The dataset spans from January 1, 2018, to September 26, 2020, with the initial 73% allocated to the Training phase and the remaining 27% reserved for Validation. The data, presented as Time Series (TS), is modeled utilizing the Autoregressive Integrated Moving Average (ARIMA) approach for prediction. Correlation analyses are conducted on the data, and ARIMA (p,d,q) parameters are determined using the Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF). The resultant ARIMA-based models for the respective MNOs are: ARIMA (0,1,3) for MNO W, ARIMA (1,0,1) for MNO X, ARIMA (2,0,4) for MNO Y, and ARIMA (0,1,1) for MNO Z. Predictive models are subsequently employed to forecast BTS Availability for the MNOs between September 27, 2020, and December 20, 2020. Model performance is assessed during the validation phase using metrics such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). MAE scores are recorded as follows: MNO W (1.3959), MNO X (0.6602), MNO Y (1.5666), and MNO Z (0.6177), while corresponding MAPE values are: MNO W (0.0150), MNO X (0.0068), MNO Y (0.0176), and MNO Z (0.0063). Additionally, Long Short-Term Memory (LSTM) models are employed for comparison against the ARIMA models for the same MNOs, yielding MAE and MAPE results of: MNO W (2.8397, 0.0322), MNO X (0.8894, 0.0092), MNO Y (2.8223, 0.0349), and MNO Z (1.1245, 0.0118). Comparative analysis reveals that LSTM models exhibit higher MAE and MAPE values than ARIMA models by percentages of 51%, 26%, 44%, and 45% for MNO W, MNO X, MNO Y, and MNO Z, respectively. This pattern remains consistent for the respective MAPE values, where LSTM models display higher values by 53%, 26%, 50%, and 47% for the same MNOs. These outcomes underscore the superior performance of the ARIMA models across all MNOs. The minimal MAE and MAPE values generated by the predictive models suggest their close alignment with actual values, facilitating effective planning and decision-making. Notably, the developed Predictive Maintenance (PdM) algorithm enables MNOs to proactively schedule maintenance activities. This algorithm, guided by a 95% availability threshold, demonstrates that MNOs W and Y yield no maintenance count savings, while MNOs X and Z experience savings of 33% and 32%, respectively.

Creating an Autoregressive Integrated Moving Average-Driven Predictive Model for Anticipating Base Station Availability, GET MORE, ACTUARIAL SCIENCE PROJECT TOPICS AND MATERIALS