Machine Learning Practical Course: Build 30 Projects

Create an image captioning project that uses Flickr data, data cleaning, and preprocessing for images and captions, applies word embeddings and transfer learning with Net50, and evaluates with Blue Score.

Convert captions to lowercase and remove punctuation, numbers, and short words. Build vocabulary from the cleaned text and keep only frequent words above a threshold of ten, storing cleaned descriptions.

Load and split training and testing data, create the train description dictionary with start and end sequences, and extract image features using the resonant 50 model for caption generation.

Feed data in batches with a generator to predict image captions from image features and a partial caption, using a 50-dim glove embedding matrix in a merged two-part model.

Train an image captioning model using categorical cross entropy and atom optimizer, with ten epochs and small batches, save the best model, and evaluate captions with BLEU scores.

Load libraries and datasets for a Costa Rican poverty level prediction. Prepare a notebook, inspect the training and test CSVs, and remove null values to clean the 995,579-row, 143-column dataset.

Master data preprocessing and feature engineering by handling nulls, encoding categoricals, dropping useless columns, and creating features like rent per person and per bedroom to improve model accuracy.

Load and compare classifiers like lightgbm and catboost, apply bayesian optimization and cross-validation for hyperparameter tuning, and use stacking classifiers to boost accuracy in predictions.

Build a stroke prediction project by loading, cleaning null values, and visualizing correlations in a dataset, then train and compare classification models using age, gender, hypertension, heart disease, glucose, BMI.

Perform data cleaning by identifying and removing null values, then explore distributions and correlations, including gender and smoking status, before encoding string features with label encoding for model readiness.

Prepare the model after cleaning data, split into train and test sets, and compare decision tree, random forest, and neural networks, achieving around 98% accuracy.

Apply principal component analysis for linear dimensionality reduction, compare PCA before/after with decision tree, random forest, and MLP classifiers, using label encoding and cross-validated accuracy to predict stroke.

Import and prepare data for a car price prediction project by loading the dataset with Pandas and NumPy, identifying and encoding categorical features, and dropping the car name column.

Transform a car dataset by creating a 'number of years' feature, apply one-hot encoding, visualize correlations with a heatmap, and identify top features predicting selling price using ExtraTreesRegressor.

Split data 70/30, compare random forest, XGBoost, and CatBoost regressors using randomized search CV and grid search CV, evaluate with MAE, MSE, RMSE, and identify the best model.

Build a Big Mart sales prediction model by loading a dataset of 1559 products across ten stores, applying regression techniques, and preprocessing categorical features with pandas, seaborn, and mad lib.

Identify null values, remove unused identifiers, and impute missing data with mean or mode; map item prefixes to food, drinks, or non consumable and compute outlet years.

Perform exploratory data analysis to visualize item distributions, apply log transformation to outlet sales, assess correlations, and encode categorical features with label and one-hot encoding.

Drop unused ids, set item outlet sales as target, and split 70/30; compare regressors (linear, lasso, random forest, lgb, boosting) and identify outlet type zero and item mrp as predictors.

perform hyperparameter tuning using randomized cross-validation across models such as random forest regressor and LGBM regressor, identifying best parameter sets and showing improved negative mean squared error.

This lecture introduces loan approval prediction project, importing pandas and seaborn, exploring a 614-row dataset of 13 features including loan status, and preparing to convert categorical columns to numerical features.

Learn practical data preprocessing and visualization for loan data: fill nulls, apply log transforms, create total income, build a correlation heat map, and encode categoricals for modeling.

Separate X and Y from the dataset, use all features except zone status to predict loan status, split 70/30 for training and testing, and compare multiple classifiers with fivefold cross-validation.

Demonstrates hypertuning models using randomized cross-validation and a random grid to optimize random forest, decision tree, and other classifiers with tuned max depth and max features for about 80% accuracy.

Predict employee attrition with machine learning by exploring a dataset of 25,491 rows and ten variables, performing feature selection, fitting multiple models, and validating with tenfold cross-validation.

Explore data types, null values, and key statistics for a 25,500-row dataset, then convert department and salary to numerical features via one-hot encoding, prepare X and y, and visualize turnover.

Apply recursive feature elimination to select top features, then train and compare logistic regression, random forest, and SVM, validating with tenfold cross-validation for robust accuracy.

Tune hyperparameters of a random forest using randomized search cross-validation, comparing accuracies across classifiers; achieve about 96.54% with tuned parameters, and review feature importance.

Predict hotel booking cancellations by loading the dataset, performing exploratory data analysis, and building and comparing models with a pipeline, train-test split, scaling, and cross-validation.

Learn practical data preprocessing and exploratory data analysis for hotel booking data, including cleaning nulls, encoding categorical data, and deriving insights from plots and correlations.

Explore data preprocessing and feature engineering with pipelines that separate numerical and categorical features, drop non-generalizable columns, and compare random forest, decision tree, and logistic regression through cross-validation.

Import numpy, pandas, and sklearn to build a temperature and apparent temperature predictor from hourly weather data, using humidity, wind speed, visibility, precipitation type, and daily summary.

Remove unused columns, convert dates, and clean missing values, then explore correlations and visualize temperature, humidity, wind, and pressure to predict temperature before a 70/30 train-test split.

Compare linear regression with XGBoost, LightGBM, and CatBoost on predicting apparent temperature from temperature, humidity, wind speed, and pressure, evaluating mae and rmse and exploring stacking and hyperparameter tuning.

In the machine learning practical course, build and evaluate regression models from linear regression to boosting and ensemble regressors. Clean data, analyze correlations, and compare model performance to improve accuracy.