Modelling of enhanced predictive analytics and decision-making approach using recurrent autoencoder with attribute subset reduction for students’ academic performance

Education plays a vital part in society as it helps economic growth through human capital, decreases crime, and enhances common well-being. Currently, predictive modelling plays a crucial role in decision-making procedures in each branch of action. Conventional techniques for recognising at-risk students often rely on reactive events that occur only after academic complexities have already been demonstrated, usually too late to avert failure or withdrawal. The development of predictive analytics powered by artificial intelligence (AI) provides a paradigm shift in proactive intervention tactics that can recognise students at risk of academic failure or dropout before vital thresholds are attained. Predictive analytics has developed as a transformative method in educational technology, leveraging AI and machine learning (ML) models to recognise at-risk students, forecast academic performances, and recommend targeted interventions before failure happens. ML for predictive analytics is altering data-driven decision-making through industries by leveraging enormous datasets and innovative techniques to discover hidden patterns and estimate future tendencies. In this manuscript, a Predictive Analytics and Decision-Making Using Recurrent Autoencoder with Dimensionality Reduction (PADM-RAEDR) approach is presented in the education sector. The aim is to develop an effective framework that enhances predictive analytics and supports effective decision-making in the education domain. At first, the data pre-processing stage is done by applying Z-score standardisation. For an effective feature selection process, the PADM-RAEDR model employs mutual information (MI), symmetric uncertainty (SU), and minimum redundancy maximum relevance (mRMR) to remove irrelevant and redundant data. At last, the long short-term memory with auto-encoder (LSTM-AE) method is employed for classification. The comparison analysis of the PADM-RAEDR model demonstrated a superior accuracy value of 98.61% over existing methods under a benchmark dataset.