Research On Botnet Detection Model Based On Random Forest And Denoising Auto-encoder

Botnets have become a very serious threat in the field of cybersecurity.They have proven to be capable of infecting billions of computers and driving them to illegal activities,causing huge economic losses to society.Effective detection of botnets is expected to alleviate this phenomenon and help to curb cybercrime.In this paper,a lot of researches on botnet detection models based on random forest and denoising auto-encoder are carried out,mainly from two aspects: model classifier and automatic feature engineering.At present,due to the huge amount of network traffic data,many botnet detection methods based on machine learning have emerged.But most methods only include a single learner,which makes the model less robust.At the same time,the relevant models rely heavily on manual feature engineering in the construction process.The consequence is that it requires a lot of expertise,a long time to build the model,and an unsatisfactory detection of the botnet.In response to the above problems,this paper has done the following work:1)A botnet detection model based on random forest is proposed.In the classifier part of the model,a random forest method based on ensemble learning is selected,which contains a large number of base classifiers.In the model training process,each base classifier learns different parameters and describes the original data from multiple angles,which effectively alleviates the instability of the detected results brought by the single model and improves the detection effect of the botnet.2)A botnet detection model based on denoising auto-encoder is proposed.The denoising auto-encoder is used to extract the original data automatically,which saves the human resources of the manual feature engineering and reduces the construction cost of the botnet detection model.The advanced abstract features extracted by the denoising auto-encoder help to improve the effect of the base classifier on botnet detection.