Matthews correlation coefficient (MCC) evaluates binary classifications with one balanced score, even when the classes differ greatly. It uses true positives, false positives, false negatives, and true negatives: +1 means perfect prediction, 0 means no better than random, and −1 means total disagreement. section 5.5.5
Customer-review classification
For review sentiment, compare keyword-based positive predictions with observed positives defined by review scores of 4 or 5. Derive the four confusion-matrix counts, then use MCC to summarize classification quality and compare models. Validate quality with a small manually labeled dataset rather than automatically labeled examples. section 5.5.5
