Maintaining road safety and ensuring vehicle integrity are critical concerns that necessitate accurate detection and categorization of road hazards, such as potholes. This paper introduces a novel technique for pothole detection using accelerometer data collected from a Raspberry Pi 4 equipped with an MPU-6050 sensor. By analyzing multi-axis acceleration data, the proposed method differentiates between normal road conditions and those affected by potholes or bumps. Six machine learning classifiers—Support Vector Machine (SVM), Logistic Regression, Gaussian Naive Bayes, K-Nearest Neighbors (K-NN), Decision Tree, and Random Forest—were individually trained and evaluated for their effectiveness in recognizing road conditions. Furthermore, a stacked generalization ensemble technique was applied to harness the strengths of these classifiers, thereby improving overall classification performance. The ensemble model significantly outperformed the Gaussian Naive Bayes, K-NN, Decision Tree, and Random Forest classifiers, achieving superior results in classification accuracy. Specifically, SVM achieved the highest accuracy at 95.21%, followed by Logistic Regression (95.08%), Gaussian Naive Bayes (94.79%), K-NN (94.56%), Decision Tree (93.10%), and Random Forest (93.12%). A meta-model was developed through stacked generalization, which achieved an accuracy of 94.98%, surpassing several individual classifiers. This research contributes to the development of advanced road condition monitoring systems, which hold the potential to enhance road maintenance and safety.