Deceptive reviews on e-commerce platforms under-mine consumer trust and distort market competition. While prior work has predominantly relied on lexical bag-of-words representations such as Term Frequency–Inverse Document Frequency (TF-IDF) in isolation, such features fail to capture writing-style irregularities and sentiment-rating inconsistencies that are strong behavioral indicators of deception. This paper proposes a Hybrid Feature Fusion (HFF) framework that constructs a unified review representation by concatenating four complementary feature groups: (i) TF-IDF weighted bigram features encoding lexical content, (ii) a six-dimensional stylometric vector capturing writing-style signatures, (iii) a novel Sentiment-Rating Consistency (SRC) score that quantifies the alignment between the polarity of review text and its accompanying numeric star rating, and (iv) surface metadata features including review length and punctuation statistics. Three classifiers—Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF)—are trained on both the TF-IDF baseline and the full HFF representation, enabling a systematic ablation study. Experiments on balanced corpora drawn from the Amazon Customer Reviews and Yelp Open Datasets demonstrate that RF trained on HFF features achieves 93.4% accuracy and an F1-score of 0.934, outperforming the TF-IDF-only RF baseline by 6.3 percentage points. Ablation results confirm that the SRC score and stylometric features provide complementary discriminative signals beyond lexical content alone. The proposed system is computationally lightweight, fully reproducible, and suitable for real-time integration into e-commerce platforms.