A framework towards more accurate and explanatory student performance model

Abstract

The ability to predict students academic performance in a timely manner is very important in learning institutions [18]. Student performance prediction is an important area as it can help teachers identify students who need additional academic support [14]. Predicting student academic performance helps teachers develop a good understanding of how well or how badly students are doing in their classes so that teachers can take proactive steps to improve student learning. Accurately predicting students future performance based on their ongoing academic records is critical to effectively conducting the educational interventions required to ensure students complete the course on time and in a satisfactory manner [4]. To achieve these goals, however, a large amount of student data must be analyzed and predicted using various machine learning models. Having a wealth of options is good, but deciding which model to implement in production is critical. While we have a number of performance metrics to evaluate a model, it is not advisable to implement every algorithm for every problem. This takes a lot of time and effort. Additionally, machine learning (ML) models are amazingly good at making predictions, but often cannot provide explanations for their predictions that humans can easily understand. Most machine learning-based projects focused primarily on results, on updating the accuracy of student grade models without considering mechanisms for their interpretability. Therefore it is important to know how to choose the right algorithm for a given task and how to choose the ”right” interpretability tool. To that end, we provide a guide for machine learning practitioners and researchers that shows the thought process that they might find useful in improving the performance and interpretability of their models, and they can even get great results on their prediction problems. The study concludes that machine learning practitioners can improve the performance of predictive models with data tactics such as getting more data, cleaning data, resampling data and rescaling data. And with algorithms optimization tactics by searching for the best hyper-parameters by using random search of algorithm hyper parameters to expose configurations that never thought of. And learning to combine by using a new model to learn how to best combine the predictions from multiple well-performing models. And can also improve the interpretability of the model by choosing the “right” interpretability tool. As machine learning becomes more ubiquitous, understanding how these models find answers is critical to improve their performance and reliability. When deciding to implement a machine learning model, choosing the right model mean analyzing your needs and expected results. Finally, developing the right solution to a problem in real life is rarely just an applied math problem. It requires awareness of business needs, rules and regulations, and stakeholder concerns, as well as considerable expertise. When solving a machine problem, it is crucial to be able to combine and balance them out. Those who can do this can create the greatest value.