BSBerkan Seseninsesenai.hashnode.dev·Jul 5 · 13 min readHandling Class Imbalance: Downsampling, SMOTE, and BeyondA fraud detection model scores 99.5% accuracy. Sounds excellent until you check the predictions: it labels every single transaction as legitimate. With only 0.5% of transactions being fraudulent, a cl00
BSBerkan Seseninsesenai.hashnode.dev·Jun 29 · 13 min readAIC and BIC: Choosing the Right Model Without OverfittingImagine you're fitting a curve to noisy data. A straight line misses the shape entirely, so you try a quadratic, then a cubic, then keep going. By degree 10 the curve passes through nearly every point00
BSBerkan Seseninsesenai.hashnode.dev·Jun 17 · 21 min readTrust Region Methods: From REINFORCE to TRPO to PPOIn the REINFORCE post, we built a policy gradient agent from scratch in NumPy and watched it learn CartPole. It worked — eventually. But the reward curve looked like a seismograph. One batch of unluck00
BSBerkan Seseninsesenai.hashnode.dev·Jun 12 · 13 min readLDA vs PCA: Supervised Meets Unsupervised Dimensionality ReductionYou have a high-dimensional dataset and you need to squeeze it down to two or three dimensions for visualisation or downstream modelling. The go-to move is PCA, and most of the time it works. But cons00
BSBerkan Seseninsesenai.hashnode.dev·Jun 9 · 14 min readChangepoint Detection: Finding Regime Shifts in Financial DataMarkets do not stay in one regime. The S&P 500 can cruise at 10% annualised volatility for months, then a crisis hits and volatility doubles overnight. Any model trained on the calm period is useless 00
