Social media is becoming the main source of information intake, allowing users to sharerntheir views freely and widely. However, the unregulated nature of this information accessrnis making social media platforms a ground for the proliferation of hate speech and fakernnews. It is evident that online hate speech could materialize to an offline impact beyondrnits psychological effects on victims. More particularly, for multi-nation and multireligiousrnas well as less democratic countries like Ethiopia, hate speech is causing drasticrnconsequences by triggering or igniting conflicts. Detecting hate speeches for resourcefulrnlanguages like English is getting better due to the availability of trained models andrnenough moderators. In the case of Ethiopia, except for the new declaration of hate speechrnproclamation, there are no automated hate speech detection mechanisms for the localrnlanguages, including Amharic which is the official working language of Ethiopia. One ofrnthe mitigation efforts to decrease the effect of hate speech in Ethiopia was to shut downrnInternet connections, which happened several times in the past.rnThe development of a hate speech detection system for Amharic will be a solution in manyrnaspects. 1) The system helps policymakers and peacemakers to automatically detect andrnact when hate speech comments are circulating on the Internet. 2) It also will help socialrnmedia platform owners such as Facebook and Twitter to automatically flagging haternspeech comments before it reaches larger audiences.rnEven if hate speech is a global issue, the systems which are developed for English or otherrnlanguages cannot be directly applied to detect hate speeches in Amharic. So, we need tornhave a new home-grown solution. Taking this into consideration, we developed a systemrnthat can detect and classify text into four categories. The system is developed usingrnStacked Bidirectional Long Short Term Memory Networks (SBi-LSTM) which is a varietyrnof Deep Learning based machine learning methods. This system is compared against twornof our baseline detection systems which are developed using dummy classifiers andrnclassical machine learning approaches. The deep learning system has achieved a greaterrnaccuracy result than the other systems. This deep learning system has shown a promisingrnresult by achieving a 94.8% F1-score accuracy result using fastText word embedding forrnvector representation. For the development of the system, we have collected andrnannotated 5,000 Amharic corpus data into racial, religion, gender and normal speechrncategories using our own custom annotation tool using 100 annotators.rnOur system has enabled multi-label categorical classification of hate speech which isrnuseful to get statistical information for any responsible organization to focus on thernvulnerable group, society or religion. Having a hate speech classification systemrndevelopment for Amharic is challenging due to the unavailability of an annotated dataset,rnmorphologically richness of Amharic and there was no Amharic hate speech classificationrnstudy that could be used as a baseline. Our system can be improved by having a morerndataset and by adding more other training layers in addition to the SBi-LSTM layer.