To address the challenge of limited effective data samples in lost circulation, a common yet complex risk in oil and gas drilling, this study proposes an intelligent diagnosis model for lost circulation risk based on a 1D-CNN and LSTM hybrid network. By combining the strengths of both networks, the model effectively captures spatial variation features and temporal fluctuation characteristics in time-series data, enabling efficient diagnosis of lost circulation risks. Under small sample conditions, traditional methods struggle to accurately distinguish between normal and lost circulation states, leading to unclear classification boundaries and frequent false alarms and missed detections. To overcome these limitations, this study introduces feature enhancement techniques to optimize the model's input features, significantly improving its stability and generalization capabilities. Experimental results demonstrate that the optimized model achieves improvements of 4%, 3%, and 3% in accuracy, precision, and recall, respectively, validating the significant performance gains achieved through feature enhancement and network structure optimization under small sample conditions.