Research On Cost Prediction Of Medical Academic Conferences Based On Machine Learning

With the development of Internet technology,big data,machine learning and artificial intelligence have gradually become the cutting-edge technology used to support decision-making in all walks of life.As a people's livelihood industry,the national policy for medical and health care is changing rapidly,and the management mode is also changing with each passing day.As a unique and subsistent marketing mode in the pharmaceutical industry,how to predict the event cost scientifically,reasonably and objectively has become an urgent problem for the relevant executive departments in pharmaceutical enterprises.Although many pharmaceutical enterprises have basically realized the information-based budget management,the actual management of event budget is still calculated by human.That let to many subjective factors that could make great risks and errors,and makes the efficiency of actual budget management difficult to improve.Based on the event data of Roche in 2018-2019,this paper introduces the number of hospitals at all levels and city classification to explore the relationship between the event cost and the influencing factors.The purpose of this paper is to establish a model to predict the cost of a certain academic event at the next moment and simplify the process of holding an event,so as to provide new ideas for improving the event management of pharmaceutical enterprises.Firstly,the paper introduces the background of the problem,then puts forward the purpose and significance of the research,summarizes the research results of machine learning and event budget management by scholars at home and abroad,and puts forward the starting point of the research.Secondly expound the theory of event budget management,feature engineering,GBDT,XGBoost,LightGBM and CatBoost.Then analyze the influencing factors of event budget,and carry out data preprocessing and feature engineering construction.Finally establishes the model,and obtains the final results after parameter optimization.Then compare the predicted results of XGBoost model,LightGBM model and CatBoost model with the traditional analogy estimation method.The main work includes the following aspects:(1)Data preprocessing.Fill the missing and abnormal values with the maximum possible value to ensure the capacity of the sample set.(2)Design feature engineering.Combined with the business experience,I try to extract the features that may be related to the prediction of academic events and transform the extracted features further to build a feature engineering that can simplify the process of holding an event.And I expand the variables by using dummy variable coding and Z-score standardization,sort the importance of features by using GBDT algorithm,and select the top 20 features for modeling.(3)Compared with the application of XGBoost model,LightGBM model and CatBoost model in Feature Engineering,the "Grid Search+five fold Cross Validation"method is used to optimize the parameters,and MAE,MAPE and RMSE are used as the evaluation criteria.CatBoost model has the best prediction effect among the three gradient boosting algorithms.So CatBoost model is finally used to predict the prediction interval value.Finally,the following innovative results are obtained:(1)In the process of data cleaning,combined with business experience,a variety of methods are used to deal with missing values and abnormal values,and the data quality does not decrease but increases.(2)For feature selection,combined with business experience,the city classification and the number of hospitals at all levels are introduced.Based on the traditional algorithm,the features that can simplify the process are constructed,and the most important features are selected.(3)By using four gradient boosting models and parameter adjustment optimization,the prediction model which is superior to the traditional algorithm is still obtained under the condition that the quality of the original data is not ideal.(4)The results of model comparison have important practical and guiding significance.By encapsulating the function,we can directly call the optimal prediction model,which can be used to predict the future event cost,simplify the event management process and optimize the organizational structure.