Akshay Atanure’s Post

🚀 Day 4 Complete – Real AI/ML Engineering Begins Today I learned something most beginners ignore 👇 👉 Machine Learning is NOT just about models. It’s about data preparation. 💡 In fact: 80% of ML work = Cleaning, transforming & understanding data Only 20% = Model building 🔧 What I implemented today: ✔ Data Cleaning using Pandas (handling missing values) ✔ Data Imputation (Mean & Median techniques) ✔ Feature Scaling using MinMaxScaler ✔ Exploratory Data Analysis (EDA) • Heatmap • Pairplot • Histogram • Boxplot 🐞 Real Bug I Faced: Tried saving files → got directory error Fix? 👉 Learned to handle file systems like a real developer using os.makedirs() 🧠 Key Insight: Bad data = Bad model Clean data = Powerful predictions 📊 Biggest Learning: Visualization helped me see patterns instead of guessing them ✔ Experience strongly impacts Salary ✔ All features showed positive correlation ✔ Dataset was clean with no major outliers 🚀 This journey is changing my mindset: From writing code ➡ to thinking like an engineer #AI #MachineLearning #DataScience #LearningInPublic #Python #GitHub #EDA #100DaysOfCode #TechJourney

"Feature engineering is where the magic happens in production ML models, yet it's often overlooked as just a preliminary step." As a data scientist, I've found that the right features can make or break your model's performance. Good feature engineering starts with understanding the data's context and business need. Here’s a simple yet effective Python snippet demonstrating how to create interaction features that capture non-linear relationships using pandas: ```python import pandas as pd # Assume df is your DataFrame df['interaction_feature'] = df['feature1'] * df['feature2'] # Scale the new feature for better model performance from sklearn.preprocessing import StandardScaler scaler = StandardScaler() df['interaction_feature_scaled'] = scaler.fit_transform(df[['interaction_feature']]) ``` This snippet shows how a simple interaction between two features can add significant predictive power. But it’s more than just creating features—it's about iteration, testing, and refining. In my workflow, leveraging AI-assisted development has transformed how quickly I can iterate through feature sets, testing hypotheses in minutes rather than hours. How do you approach feature engineering in your projects? Any tips or tricks you'd like to share? #DataScience #DataEngineering #BigData

🚨 I thought my ML model was broken… Turns out, my data was lying to me. Last week, I was building a customer segmentation pipeline. Everything looked fine — clean dataset, logical features, decent approach. And then… chaos. Random errors. Broken calculations. Features behaving in ways that made ZERO sense. After hours of debugging, I realized: 👉 The problem wasn’t my model. 👉 It wasn’t even my logic. 👉 It was my assumptions about the data. Here are some mistakes that completely humbled me 👇 🔴 “It looks numeric” ≠ It is numeric 0,1,2 sitting in a column… but dtype = object → Boom: math operations fail 🔴 Datetime betrayal "21-08-2013" Pandas: “Month = 21? I’m out.” 🔴 .replace() illusion I encoded categories… but forgot that dtype stays object 🔴 The silent bug in drop() Used axis + columns together → Pandas said: “choose one bro” 🔴 Fake logic: “< 25 unique = discrete” Worked… until it didn’t 🔴 Redundant features everywhere Created multiple columns… doing the SAME thing 🤦♂️ 💡 Biggest lesson: Most ML problems are not model problems. They are data understanding problems. Now, before touching any model, I ALWAYS check: ✔ df.info() ✔ df.dtypes ✔ hidden type issues ✔ assumptions vs reality This debugging session changed how I approach ML. Less focus on fancy models. More focus on respecting the data. If you’re learning ML right now, remember this: 👉 The model is the easy part. 👉 Data is where the real game is. Curious — what’s a bug that completely fooled you at first? 👇 #MachineLearning #DataScience #Python #Pandas #LearningInPublic #AI

🚀 Really excited to share this insightful study we completed for the Theory and Practices in Statistical Modelling (IT3011) module at SLIIT! In this project, we explored a very practical and often overlooked problem in machine learning: 🔺 How does noise in data affect model performance and robustness? Working with the Red Wine Quality dataset, we experimented by injecting different types of noise (Gaussian noise, missing values, and outliers) and evaluated how models like Linear Regression, Random Forest, and SVR responded. 📊 One key takeaway for me: Even the best models can fail if the data quality is poor. This project really showed that data quality is just as important as model selection. 💡 It was also a great experience building an interactive dashboard to visualize how model performance degrades with increasing noise levels. Big thanks to our lecturer Samadhi Chathuranga Rathnayake for the continuous guidance, and to my amazing teammates for the teamwork! 🔗 Check out the full project and interactive dashboard here: https://lnkd.in/gGwJYdeB #DataScience #MachineLearning #DataQuality #SLIIT #LearningJourney #AI #DataAnalytics

Machine Learning/Artificial Intelligence  Day 12.  Today, I worked on a large sales dataset and ran 7 different analyses to uncover hidden patterns.What I did:First, I loaded the dataset into Jupyter using pandas. The data had thousands of rows with sales records across different regions, products, and shipping methods.Then I asked specific questions:1. Which region made the most sales?2. Which product sold the highest quantity?3. Which ship mode had the most delays?4. How do sales trend across different months?5. Which product category brings in the most revenue?6. Is there a relationship between discount and profit?7. Which region prefers which ship mode?Tools I used:· pandas – to clean, filter, and group the data· seaborn & matplotlib – to create histograms, bar charts, pie charts, and line graphs· Jupyter – for writing and testing my code· Google Colab – to share the notebook and collaborate· GitHub – to save, track, and share my workWhat I found:One region alone made up 40% of total sales. One product sold three times more than others. And the fastest ship mode actually had the most late deliveries – a surprising insight.Why this matters:For AI/ML, understanding your data before building models is half the work. A good chart can save hours of wrong assumptions. And sharing work on GitHub keeps everything organized and open for feedback.What I learned today:EDA is not just about making charts. It is about asking the right questions. Visualization is not just about pretty colors. It is about telling a clear story. Collaboration is not just about sharing files. It is about making your work useful to others.Learning step by step, staying consistent every day!#M4ACE LearningChallenge#LearningInPublic#30DaysOfAIML#EDA #DataVisualization #Python #pandas #seaborn #matplotlib #GitHub

𝗗𝗮𝘆 𝟭𝟯 𝗼𝗳 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗔𝗜/𝗠𝗟 🚀 Today I dove into data preprocessing — specifically centering and scaling, one of the most impactful steps before training a model. 𝗞𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀: Why it matters: Features with wildly different ranges (like duration in milliseconds vs. speechiness as a decimal) can bias models that rely on distance — like KNN — making scaling essential. 𝗠𝗲𝘁𝗵𝗼𝗱𝘀 𝗰𝗼𝘃𝗲𝗿𝗲𝗱:  • 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗮𝘁𝗶𝗼𝗻 — subtract the mean, divide by variance → zero mean, unit variance  • 𝗠𝗶𝗻-𝗠𝗮𝘅 𝗡𝗼𝗿𝗺𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 — scales data to [0, 1]  • 𝗖𝗲𝗻𝘁𝗲𝗿𝗶𝗻𝗴 — scales data to [-1, 1] 𝗪𝗵𝗮𝘁 𝗜 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗲𝗱 𝗶𝗻 𝘀𝗰𝗶𝗸𝗶𝘁-𝗹𝗲𝗮𝗿𝗻: Using StandardScaler from sklearn.preprocessing Applying fit_transform on training data and transform on test data (to prevent data leakage!) Building a Pipeline that chains scaling + KNN together cleanly Combining GridSearchCV with a pipeline for tuned cross-validation 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁 𝘁𝗵𝗮𝘁 𝗯𝗹𝗲𝘄 𝗺𝘆 𝗺𝗶𝗻𝗱: KNN on unscaled data → 53% accuracy. KNN on scaled data → 81% accuracy. That's a 50%+ boost 𝗷𝘂𝘀𝘁 𝗳𝗿𝗼𝗺 𝗽𝗿𝗲𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴! 🤯 Small steps, big impact. Preprocessing isn't glamorous, but it's where good models are made. #100DaysOfML #MachineLearning #DataScience #ScikitLearn #Python #AI #LearningInPublic #Day13

Why do customers leave? Let's ask the data. Project 1, Day 1: Data Engineering & EDA for Customer Retention. I just kicked off a new Advanced AI project: A Churn Prediction Pipeline. It costs 5x more to acquire a new customer than to keep an existing one, making churn prediction one of the most valuable ML applications in business. But before I can train any AI, I need clean data. Real-world databases are messy. Today, I built a Data Engineering dashboard using Python, Pandas, and Streamlit to: ✅ Clean invalid datatypes and handle missing values (Imputation). ✅Perform Exploratory Data Analysis (EDA) to find visual trends. ✅Apply One-Hot and Binary Encoding to translate text into numbers for the algorithm. The biggest insight from the EDA? Month-to-month contracts are the massive driving force behind churn, while long-term tenure customers rarely leave. Now that the data is mathematically clean and encoded, it's ready for the AI. Tomorrow: Training the XGBoost algorithm to mathematically predict exactly who is going to cancel next! #Python #DataEngineering #DataScience #MachineLearning #CustomerRetention #Streamlit #Analytics

Building a Machine Learning Model for Time Series Forecasting Over the past few days, I’ve been working on a machine learning project focused on predicting future values using real-world financial data. 🔍 What I worked on: Data collection and preprocessing using pandas Feature engineering and handling missing values Implementing regression models such as Linear Regression Training and evaluating models using scikit-learn Using historical data to forecast future trends Visualizing predictions with matplotlib 📊 Key Techniques Applied: Data cleaning and transformation Train-test splitting Model training and evaluation Time series forecasting using shifted labels Scaling features for better model performance 📈 What I achieved: Built a working model that predicts future values based on historical patterns Compared actual vs predicted results using visual plots Gained deeper understanding of how machine learning models learn from data 💡 Key takeaway: Machine learning is not just about building models—it’s about understanding data, preparing it properly, and interpreting results effectively. 🎯 Next steps: Improve model accuracy with advanced techniques Explore additional models and comparisons Build more real-world projects and expand my portfolio I’m excited to continue growing in Data Science and Machine Learning and apply these skills to real-world problems. #MachineLearning #DataScience #Python #AI #DataAnalysis #LearningJourney

🚀 Day 39 of My Data Science And Machine Learning Journey 👉 ColumnTransformer + Pipeline + GridSearchCV + Logistic Regression Today I implemented a complete ML workflow using Scikit-learn — something that’s actually used in real-world projects. 🔧 What I built: ✅ ColumnTransformer → Handles different data types (numerical + categorical) ✅ Pipeline → Connects preprocessing + model into one flow ✅ GridSearchCV → Finds the best hyperparameters automatically ✅ Logistic Regression → Final model for prediction 🧠 Key Learning Instead of writing separate code for: preprocessing ❌ training ❌ tuning ❌ 👉 I combined everything into ONE clean pipeline ✅ 🔥 Why this matters ✔️ Prevents data leakage ✔️ Makes code reusable ✔️ Ensures consistency in training & testing ✔️ Industry-level best practice 💡 What it does: Loaded dataset Applied preprocessing using ColumnTransformer Built Pipeline Tuned model using GridSearchCV Evaluated performance 📌 This is how real ML systems are built — not just models, but complete workflows. #MachineLearning #DataScience #AI #Python #ScikitLearn #MLPipeline #FeatureEngineering #LearningInPublic 🚀

Day 23: Real-World Data Ingestion & Feature Extraction in Pandas 🐍🤖 To build autonomous Agents and robust RAG pipelines, you need a flawless data foundation. Today, I completed my Pandas deep dive, shifting away from theory and executing end-to-end data extraction on messy, real-world datasets. Here are the core engineering takeaways from the final projects: 🌍 Real-World Data Ingestion: Handled importing and profiling massive .csv datasets. In Generative AI, this is step zero. Before an LLM can process a document, the raw data must be loaded and structured cleanly into memory. 🧩 Advanced Feature Extraction: Applied custom Python functions across unstructured text columns to parse hidden variables and generate brand-new, clean data points. This is exactly how you generate high-quality metadata to enrich documents before feeding them into a Vector Database. 🔎 Precision Querying: Chaining operations like .loc, .nlargest(), and conditional masking to extract highly specific insights. When building Agentic AI, writing this backend logic is how you give an Agent a functional "Database Search" tool. With NumPy (matrix math) and Pandas (data wrangling) officially locked in, the computational architecture is set. It is finally time to start building the "brain". #Python #GenAI #AgenticAI #MachineLearning #Pandas #LangChain #DataEngineering #100DaysOfCode