Advanced Traffic Forecasting Integrating Temporal and Spatial Dependencies Using Hybrid Deep Learning Models

Accurate traffic forecasting is important for refining traffic management and planning to avoid congestion on the roads and enhance road safety. Traditional models often misfire on complex, nonlinear patterns in traffic. In this study, the hybrid LSTM-CNN model proposed in this paper would overcome the limitations by modeling both temporal and spatial dependencies, thus ensuring better accuracy and reliability in prediction. The study portrays the hybrid model of LSTM-CNN to overcome all such limitations and focus on capturing the temporal and spatial dependencies pertaining to traffic features. The paper uses a rich dataset comprising variables like volume, speed, and occupancy from highway sensors. It gives a model using LSTM layers in combination with CNN to perform better in prediction. Further refinements were done in training using hyperparameters; the evaluation of performance was executed on R², MAPE, and RMSE. The hybrid model gave the lowest validation loss of 0.05 and the lowest test MAPE of 0.08, which is better than the conventional models. More precisely, from the LSTM model, R² score = 0.081, MAPE = 3.66%, and RMSE = 0.248; from the CNN model, R² score = 0.029, MAPE = 4.07%, and RMSE = 0.255. R² of 0.063, MAPE of 3.84%, and RMSE of 0.250 were found for the hybrid model, with LSTM before CNN. In reversed order—that is, the hybrid model of CNN first—the values are as follows: the model recorded an R² of 0.054, a MAPE of 4.15%, and an RMSE of 0.252.