Accurate construction cost estimation is essential for efficient project planning, particularly in developing countries like Bangladesh, where cost overruns are common. This study applies two simple machine learning algorithms, Decision Tree Regressor and K-Nearest Neighbors (KNN), to predict final construction costs of residential buildings using a dataset of 50 completed projects from various regions and seasons in Bangladesh. Key features include floor area, number of floors, construction duration, soil and foundation types, roof type, and cost components. The data were collected through structured field surveys and contractor records and preprocessed using standardization and one-hot encoding. Model performance was evaluated using MAE, RMSE, and R² score. The KNN model outperformed the Decision Tree, achieving an MAE of 0.276 million BDT, RMSE of 0.370 million BDT, and R² of 0.60. In contrast, the Decision Tree model yielded an MAE of 0.341 million BDT, RMSE of 0.442 million BDT, and R² of 0.44. An R² of 0.60 indicates a moderate but reliable level of predictive power that avoids overfitting, especially valuable in small, real-world datasets. Feature importance analysis from the Decision Tree revealed that material cost (62.1%), location (21.3%), and labor cost (12.5%) were the most significant predictors. The findings suggest that interpretable and lightweight models can support cost forecasting even in data-scarce environments, aiding better budgeting and resource allocation in residential construction.