1. Data Leakage during Preprocessing
Doing preprocessing (like scaling or imputation) on the whole dataset before splitting into train/test leaks test information into the training phase.
2. Using Accuracy for Imbalanced Data
If 99% of your data is negative, a dummy model predicting always negative gets 99% accuracy but is useless. Use Precision, Recall, or F1-score instead.
3. Ignoring Baseline Models
Before jumping to deep neural networks, always build a simple baseline (like Logistic Regression or a simple Decision Tree).