Strategic planning for food security has become the key intention especially for the developing countries like Pakistan. A comparative study is carried out to forecast the wheat area, yield and population eruption in Pakistan using the time series dataset, comprise from 1950-2020.This study layouts the plan to develop the regression model using the compound growth rate, called compound growth exponential regression models (CGREM). CGREM are applied using the machine learning (ML) and statistical approaches to address the food security planning for wheat area, yield and population eruption in Pakistan. Data partition is carried out using 80% and 20% randomized partitions for ML models. The hyper parametric tuning of ML model is further applied for 75%, 25% and 70%, 30% randomized partitions. The Performance of ML models are evaluated based on training and testing datasets. The evaluation metrics (RMSE, R2) and information criterions (AIC, SIC) are used to measure the performance of models. The decomposition prediction error (P.E) is used to address the variance bias tradeoff and to select the optimum model. The decomposition model is applied to decompose the wheat production into its determinants. CGREM found better fitted model using the machine learning approaches. CGREM predicted, up to 2050, wheat area will rise up to 51.7%, wheat yield will grow up to 109.7%, and population will rise up to 140.6%. It noted that population will likely to upturn 88.9% and 30.9% more from wheat area and yield. Decomposition analysis model depicts that wheat productivity and area sharing 38% and 20% change towards wheat production. This study demonstrated the strong evidences to layout the true policy decisions, which leads to overcome the social dilemma of food security.