A Time Series Analysis of the Growth and Development of Higher Education in Nepal

Abstract

Nepal’s higher education sector has expanded substantially, highlighting the importance of enrollment forecasting for capacity planning and quality assurance. This study analyzed student enrollment in Nepal’s universities and academies from fiscal year 2014/15 to 2024/25 and compared two forecasting approaches. Data were compiled from University Grants Commission (UGC) annual reports, the Education Management Information System Report on Higher Education 2023/24, and contemporaneous press reports citing UGC statistics. The analysis applied a Box–Jenkins autoregressive integrated moving average (ARIMA) model and a long short-term memory (LSTM) recurrent neural network implemented from first principles and trained using backpropagation through time. Augmented Dickey–Fuller testing indicated that the enrollment series was nonstationary in levels but stationary after first differencing. An ARIMA(1,1,0) model was estimated using ordinary least squares, yielding Δyₜ = 24,203.6 + 0.581Δyₜ₋₁. The ARIMA model achieved a lower in-sample mean absolute percentage error (5.75%) than the comparably sized LSTM network (8.42%). Both models projected continued enrollment growth, with ARIMA projections indicating an increase from approximately 749,000 students in 2024/25 toward 825,000–955,000 by 2027/28. These projections were contextualized by a gross enrollment ratio of 22%, higher female than male participation (24.6% versus 17.8%), and the continued dominance of Tribhuvan University. The findings provide quantitative evidence to inform institutional capacity planning, quality assurance, and higher education policy. The methodological comparison also highlights the constraints of applying deep learning to short annual enrollment series and supports further investigation using higher-frequency, disaggregated data.