Ambient air pollution has been linked to several health endpoints. The WHO attributes 7 million deaths annually to air pollution with particulate matter (PM_2.5) being the pollutant of critical importance due to its devastating health effects. Air quality monitoring is very limited in sub-Saharan African (SSA) countries and although satellite remote sensing has helped to bridge the huge air quality data gaps, these measurements have not been validated against ground-level measurements in these countries. We therefore evaluated the efficiency of low-cost sensors in estimating PM_2.5 concentrations in an African city through comparison of low-cost sensor data with satellite aerosol optical depth (AOD) data leveraging complex machine learning (ML) methods. Low-cost sensor data were collected from a monitoring network in Accra, Ghana, with AOD measurements extracted from the MODIS MCD19A2v061 dataset and processed using the MAIAC algorithm. Ordinary Least Squares regression, Random Forest, Extra Trees, Boosted Decision Trees and XGBoost were used to establish the relationship between AOD and low-cost sensor PM_2.5 measurements incorporating meteorological data. We observed significant positive relationships for two low-cost sensors deployed in the network (Clarity Node S and Airnote). The R² values were, however, low, ranging from 0.18 to 0.27, with the corrected Airnote data recording the highest R². The ML models which integrated temperature and humidity improved the R² values with the Boosted Decision Tree demonstrating the best predictive capability. Seasonal variability was found to have a strong influence on model performances with the dry season model performing significantly better than the wet season model. Consistent with other studies, AOD explained only a small proportion of ground-level PM_2.5 variations. Evidence from this sensor network in Accra suggests that AOD predicts ground-level PM_2.5 measured with low-cost sensors in a manner similar to conventional air monitoring instrumentation. However, for low-cost sensors to be deemed a good substitute for satellite AOD, data correction with complex algorithms developed in the same research location will be required.