Temporal Trends, Inequality and Forecasting of ISP Latency in Nigeria, 2020–2023

Abstract

Internet Service Providers (ISPs) experience variability and inequality in network latency, affecting service 
quality and user experience. This study addresses assessment of latency performance by quantifying 
trends, inequality, regime changes, and forecasting behaviour across ISPs. Monthly latency measurements 
collected from January 2020 to December 2023 were cleaned by removing zero and negative observations 
and summarised using descriptive statistics. Gini and Theil indices and Lorenz curves quantified latency 
inequality, while seasonal-trend decomposition and Augmented Dickey-Fuller test examined behaviour 
and stationarity. A two-state Markov-switching autoregressive model classified network conditions into 
“good” and “bad” regimes. ARIMA, Holt-Winters, Prophet and Random Forest models were evaluated 
using CRPS and pinball loss metrics. Mean latency ranged from 8.04 ms for Dimension Data to 32.73 ms 
for InqDigital. Random Forest achieved the best performance. The study demonstrates that combining 
inequality analysis, regime classification and machine-learning forecasting can support early identification 
of elevated latency and stable ISP QoS.