The temporal variability of soil radon concentration of tectonically active origin is considered a potential earthquake precursor. However, the reliable identification of radon anomilies possibly induced by earthquake activity is still enigmatic. The present study applies intelligent algorithms such as artificial neural networks (ANN), multiple linear regressions (MLR) and decision trees (DT) for an improved and more reliable identification of radon anomalies of the tectonically active origin in northern Pakistan. The dataset utilized in the present investigation includes the soil radon concentration and associated meteorological parameters recorded at a seismically active location in northern Pakistan. The recorded dataset is subdivided into seismically active (SA) and stable (NSA) periods using a time window of ±7 days around the time of earthquake. The NSA dataset is utilized for the training/cross-validation of intelligent algorithms using three input (meteorological factors) and one output parameter (radon). The trained intelligent models are then used to simulate the expected radon level during the SA periods based on the measured environmental parameter. The discrepancy (Q) between the simulated and measured radon is utilized to identify statistically significant (x¯±2σ) radon anomalies. Our results reveal that all the employed intelligent algorithms show a significant deviation in Q values around the time of earthquake occurrence. Furthermore, comparative analysis suggests that ANN performs better than the other techniques. In conclusion, our study paves the way for the possible use of radon in designing an optimized strategy for earthquake forecasting.