This is feng shui and today my topic is identification of key influencers for secondary distribution of hiv ast among chinese msm, a machine learning approach. This study is authored by many of my colleagues and the course corresponding authors is dr ching. Phong gian, dr rimington. If you have some questions about this study, you could contact corresponding authors or contact me. First of all, i will introduce some study background some notes. This study is actually the funding and base of identification of verification of msmk influencers during secondary distribution of hiv, asd machine symbol, learning based course, experimental study, and this study has been printed by my colleague yesterday in parallelization 2 digital approach. So if you are interested in the public health background and in the domain, knowledge of hiv vst secondary distribution to influence certification, you could have some review of that presentation. And this study is the machine learning modeling part of that study, working as a training seat and the testing set of our whole study and that study works, as the predictions said, with real world course experimental validation. So this presentation will talk more about machine learning modeling with some technical details. This is the framework of this study, simply introduce this study and we used four machine learning models for project selection and for modeling and the four machine learning models are exampled as one example algorithm to final predict, who are the influencers in our hiv, st signal, distribution And to compare with our methods, we use the conventional human identification used.

The leadership safe, reported skills, cut off methods, so, as i mentioned before, this presentation will concentrate on modeling part. So what is machine learning? A good example is alphago. Africa is alphago. Is an implementation of machine learning, modeling and actually alphago used, reinforcement, learning, it’s, a very advanced, deep machine learning approach and due to the sample size of our data set and our trill? We just utilize four simple machinery models in our study: logistic regression, support vector, machine decentring and random forest and the same logistics regression in the general data analysis and the measuring modeling what’s. The difference in general data analysis of this regression is just used for analysis. The relationship of all variables y i and x i and but for machine learning modeling, some samples are used for parameters, training and the and the remaining samples are used for testing. And if you have, if we have the future x201x202, we will use the uh the the model and the x2 1×202 and so on, to calculate or to predict the value of y201 202. So the aim of this study is to provide such a predictive model for the following course experimental study. So this study served as the training and testing parts and about the primer about the results of this study. We present the machine learning classification results using five fold course validation and we would say, in sample machine learning, uh achieved the highest accuracy and fe score in three in three types of uh influencers identification key distributors refers to index who could distribute more than or equal To two kids to authors and that key detectors means indexes who could help us to find positive waters and the key promoters means that index.

Who could help us to find some newly tested authors? And we could see that the ensemble machinery achieved the highest accuracy. No matter in which types of kicking key influences identification and also if we identify the same number of influencers based on our service data, the one certain human identification happened to classify 49 influencers and in and our ensemble machine learning approach also happened to classify 49 influencers In five for testing sets and they are the same numbers, so the so search comparison is comparable and we could see that a machine learning approach could distribute more keys, could help us to find more positive waters could help us to find more newly tested orders, and We also run a simulation experiment. The simulation is just to simulate the course experimental, but by the way, based on some assumption, we can use computer programming to simulate and the simulation results is also demonstrated. The intervention efficiency is improved by our machine learning approach. So, in summary, our approach outperformed the human activation, no matter no matter in average accuracy or to find more keys, find more newly tested authors or find more positive, tested.