Drought poses a significant environmental challenge, especially in semi-arid regions, where comprehending its temporal dynamics and developing predictive models are crucial for effective water resource management. The present study investigates drought trends and forecasts standardised precipitation index (SPI) values at multiple timescales across five stations in the north dry zone of India (Bidar district: Aurad, Bhalki, Humnabad, Bidar and Basavakalyan). The Mann-Kendall test was employed for trend analysis, while auto-regressive integrated moving average (ARIMA) and artificial neural network (ANN) models were used for SPI prediction. The results revealed no statistically significant trends across all stations and timescales, though marginally significant positive trends were observed at longer timescales in Basavakalyan and Bhalki, suggesting a potential improvement in drought conditions. The seasonal auto-regressive integrated moving average (SARIMA) model effectively captured seasonality and autocorrelation, with varying autoregressive and seasonal moving average components across stations. The ANN model, structured with a '25-13-1' architecture, demonstrated superior performance during model training by capturing non-linearities, outperforming SARIMA across all statistical metrics. However, during model testing, SARIMA exhibited superior predictive accuracy with lower root mean squared error and higher Nash-Sutcliffe efficiency values compared to ANN, indicating its robustness in real-time drought forecasting. These findings underscore the importance of integrating statistical and machine learning approaches for drought prediction. While the ANN model effectively captures complex patterns, the SARIMA model remains more reliable for modelling temporal drought dynamics. The results provide valuable insights into drought dynamics, supporting informed water resource management and proactive drought mitigation strategies in the Bidar district.