Innovation In Smartphone Price Prediction Using Cross Validated Regression Models

General Background The rapidly growing smartphone market necessitates data-driven pricing models to help consumers and manufacturers estimate valuations based on complex hardware specifications. Specific Background While machine learning algorithms like Support Vector Machine (SVM) and K-Nearest Neighbors (K-NN) are widely used for product valuation, the unique heterogeneity and rapid cycles of the mobile market demand domain-specific evaluations. Knowledge Gap Existing empirical studies present mixed performance results and frequently neglect the generalization stability of these models under cross-validation scenarios. Aims This study evaluates linear SVM and K-NN models for predicting smartphone prices using technical attributes, explicitly analyzing the trade-off between predictive accuracy and model robustness. Results Utilizing a preprocessed dataset of 836 smartphone records, the K-NN model achieved lower test set errors with a Mean Absolute Error of 12,809.63 and an R² of 0.461, whereas the linear SVM recorded an R² of -0.013. However, 10-fold cross-validation exposed severe overfitting in the K-NN model with an R² of -0.659, while the SVM showed slightly better stability with an R² of 0.098. Novelty This research distinctly highlights the vulnerability of instance-based learning to local data density and skewed distributions, contrasting it with the constrained expressiveness of margin-based linear models in price prediction. Implications Future development of pricing recommendation systems must prioritize non-linear kernel exploration, robust hyperparameter optimization, and extensive cross-validation over simple error minimization. Highlights K-Nearest Neighbors regression minimizes test set errors but demonstrates severe overfitting during cross-validation. Linear Support Vector Machine provides marginally higher generalization stability despite weaker single-split predictive accuracy. Right-skewed technical specification data necessitates robust validation schemes and non-linear kernel optimization.