Machine learning-based profession field selection guide for high school students

Artificial intelligence is a system that develops by imitating people's brain and constantly renewing itself thanks to the feedback mechanism. This system is gaining a bigger place in human life day by day. The reason can be stated as when there are artificial intelligence systems in technological tools that contains software, communication with the environment is maximized with receiving feedback, and as a result, the functionality of these tools is increased. The concept of machine learning is an application area of artificial intelligence and is widely used in the industry. The most important tool that required to use machine learning is meaningful data. With this data, relationships, and results that the human eye cannot catch can be captured. With the results obtained from here, improvements can be made fort the problem domain. Educational institutions are institutions where data is available in enormous sizes. However, these data are not used much in our country yet. For this reason, the subject of this thesis was chosen in the field of education. With the help of machine learning, it has been thought that there may be a correlation between the grades that high school students have taken in courses for 4 years with some other information and the university departments they have chosen. With this motivation, the necessary permissions were obtained from the Ministry of National Education and the information of high school students who graduated in previous years were collected. A total of 447 observation units were established. The total number of departments chosen by these students was determined as 43. Since the ratio of observation units to the number of classes in the dependent variable is low, new dependent variables were created by combining similar departments. As a result, there are 4 different labeling strategies containing 43, 23, 9 and 6 classes for different experiments conducted in the study. Moreover, 10 machine learning algorithms were tested for each labeling strategy on the data set. XGBoost gives very successful results with a success rate of 43% in the first labeling strategy, 53% in the second labeling strategy, and 75% in the fourth labeling strategy. In the third labeling strategy, KNN gives the most successful result with 65%.IX In summary, it can be inferred from the study that if the observation units in the data set are increased significantly, there will be no need to create new labels by grouping the departments and the success in the predictions will increase significantly.