This article introduces an innovative approach for forecasting soil strength and state in the construction domain using deep learning and machine learning techniques. With the burgeoning demand for reliable predictive analytics in construction to ensure safety, optimize resources, and mitigate risks, our study focuses on the integration of Random Forest (RF) for feature selection and Long Short-Term Memory (LSTM) networks for dynamic soil condition prediction. We commence by collecting a diverse dataset comprising historical soil data, satellite imagery, on-site sensor readings, and weather reports. Through meticulous preprocessing and normalization, we prepare the dataset for analysis. The RF algorithm plays a pivotal role in identifying the most influential features impacting soil strength and state, streamlining the LSTM network's focus on the variables with the highest predictive power. Our LSTM model is meticulously architected to process sequences of selected features, capturing the temporal dependencies critical for accurate forecasting. The model is trained and validated on a partitioned dataset, utilizing mean squared error (MSE) for regression tasks and categorical cross-entropy for classification objectives. Rigorous evaluation on a separate test set demonstrates the model's effectiveness, showcasing its potential to revolutionize construction planning and risk management. The article not only details the technical methodology but also discusses the practical implications of deploying such a predictive analytics model in the construction industry. By leveraging the temporal pattern recognition capability of LSTM networks and the feature selection prowess of RF, we present a robust tool for forecasting soil conditions, which is vital for the preemptive planning and execution of construction projects. This study contributes to the growing body of knowledge at the intersection of construction engineering and artificial intelligence, offering a novel solution to a longstanding challenge in the field