Chimbuchi Chukwuemeka

Chimbuchi ChukwuemekaJul 14, 2026

What I'm Building Toward  Here’s where I am right now: I’ve built and evaluated a Linear Regression model on California housing prices, tuned Ridge and Lasso with cross validation, and I’m moving into Elastic Net next, the model that tries to get the best of both worlds.After that I’ll compare all four models side by side: which predicts prices most accurately, which is the most interpretable, and which actually deserves a place in a real report.What surprised me most: the “hard” part wasn’t the theory. It was discipline, reading my own code carefully, not assuming a line does what I meant, and treating every red error message as useful information instead of a personal insult.I’m not done. There’s still Elastic Net, feature importance comparisons, and a final write up to go. But for the first time in this project I feel like I’m not just running code. I understand why it works, and why it sometimes doesn’t.If you’re on a similar path, learning Python, data science, or machine learning from scratch, I’d love to hear where you are in your journey.More soon. 🚧#Python #DataScience #MachineLearning #ElasticNet #TechRiseCohort3

Chimbuchi ChukwuemekaSep 1, 2026

New week, new algorithm. This time I dived into Time Series analysis, and it completely changed how I think about data.Unlike regular datasets where you can shuffle rows freely, time series data must stay in order. Think of it like reading a novel. You cannot start from chapter 10 and expect the story to make sense. Every observation depends on what came before it.A few key things I learnedStationarity matters. A stable time series has a constant mean and variance over time. Models like ARIMA depend on this, and tests like ADF and KPSS help confirm it.Trend, seasonality, cycles, and randomness. Every time series can be broken down into these four components. Once you see them separately, forecasting becomes much clearer.Train test splitting is different here. You cannot randomly split time series data like other machine learning problems. Doing that leaks future information into training and gives false confidence in your model. The split must respect time order, using earlier data to train and later data to test.From AR and MA models to ARIMA, and now machine learning approaches like Random Forest and XGBoost for forecasting, this space is deep. Feature engineering alone, lag values, rolling averages, cyclical encoding, opened my eyes to how much thought goes into preparing time based data properly.Still early in this journey, but every concept builds on the last.If you work with time series data, whether in finance, sales forecasting, or business analytics, and need someone eager to learn and apply these methods, I would love to connect and collaborate.#MachineLearning #TimeSeriesAnalysis #DataScience #AI #Python #TechRise #ARIMA #Forecasting"

Chimbuchi ChukwuemekaSep 1, 2026

"New algorithm unlocked. This week I studied Support Vector Machines (SVM), and the core idea is beautifully simple.SVM does not just try to separate two classes. It tries to separate them with the widest possible gap. That gap is called the margin, and the closest points to it, the ones that actually decide where the boundary sits, are called support vectors.A few key concepts that stood out to me:Hard Margin vs Soft Margin. Hard Margin SVM assumes your data is perfectly separable, no noise, no overlap. In the real world, that almost never happens. Soft Margin SVM allows a few misclassifications using slack variables, controlled by a parameter C. A large C punishes errors heavily and risks overfitting. A small C allows more flexibility and often generalizes better.Linear vs Non Linear SVM. Not every dataset can be split with a straight line. Non Linear SVM solves this using kernel functions, especially the RBF kernel, which projects data into a higher dimension where it becomes separable. There is also a gamma parameter here. Low gamma gives a smoother, simpler boundary. High gamma fits the training data tightly but risks overfitting.Applications are everywhere: spam detection, sentiment analysis, fraud detection, face recognition, medical diagnosis, even bioinformatics.What I appreciate most about SVM is the philosophy behind it. It does not just find a boundary, it finds the most confident one.Learning one algorithm at a time, one lecture at a time. If you are working on classification problems and need someone who understands both the math and the code behind it, let us connect.#MachineLearning #SVM #DataScience #AI #Python #TechRise #SupportVectorMachines"