There is an increase in vehicle use in Addis Ababa which increases by 5% annually, not only this but also the urban arterial roads are of mixed type and there are no automated means that monitor the road traffic congestion. This creates road traffic congestion which is one of the main problems that affects everyone in time, energy and resource management. Road traffic estimation is instrumental in developing a more advanced transportation system to avoid congestions, trip delay, improper fuel usage, and vehicle use. There are different methods to collect data and estimate the degree of road traffic congestion such as manual counting (observation), global positioning system, inductive loops, using a camera system. The latter three mentioned are expensive in installation and maintenance with coverage limitations. This thesis focuses on one of the cost-effective alternative methods, which is based on 3G cellular handover data in the case of Addis Ababa's main selected road for the experiment. In this study, we examined this alternative method to estimate the degree of road traffic congestion using a simple feedforward backpropagation neural network by considering and ignoring the altitude parameter. Related research has been done, but this study does not show by considering the altitude as a factor. We collect handover data while driving along selected arterial roads in Addis Ababa, and also handover archived data has been used for some sections of the streets for the experiment. For the congestion estimation data classification, AADT(Annual average daily traffic) data has been collected from AAPC(Addis Ababa police commission), and LOS(level of services) also used, which is based on the speed of test vehicles. The neural network was then trained and tested using the collected data against the road level classification done. The results found showed better performance of congestion estimation with an accuracy of 92.5% when the feature altitude is considered whereas without considering this variable, the result found is similar to those of other related research works 82.1 %. It has been shown from the effects that the improvement in the accuracy of estimation increases by 10%.