Logistic Regression<\/strong><\/p>\n\n\n\nUsed for binary classification, predicting one of two outcomes from input features.<\/p>\n\n\n\n
from sklearn.linear_model import LogisticRegression\nfrom sklearn.model_selection import train_test_split\n\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\nmodel = LogisticRegression()\nmodel.fit(X_train, y_train)\npredictions = model.predict(X_test)\nprint(\"Accuracy:\", model.score(X_test, y_test))<\/code><\/pre>\n\n\n\n2. Random Forest<\/strong><\/p>\n\n\n\nAn ensemble of decision trees that reduces overfitting compared to a single tree.<\/p>\n\n\n\n
from sklearn.ensemble import RandomForestClassifier\n\nrf_model = RandomForestClassifier(n_estimators=100, random_state=42)\nrf_model.fit(X_train, y_train)\npredictions = rf_model.predict(X_test)\nprint(\"Feature importances:\", rf_model.feature_importances_)<\/code><\/pre>\n\n\n\n3. Forecasting Models (ARIMA)<\/strong><\/p>\n\n\n\nForecasts future values in a time series based on its own past values.<\/p>\n\n\n\n
from statsmodels.tsa.arima.model import ARIMA\n\nmodel = ARIMA(sales_series, order=(1, 1, 1))\nfitted_model = model.fit()\nforecast = fitted_model.forecast(steps=12)\nprint(forecast)<\/code><\/pre>\n\n\n\n4. Clustering (K-Means)<\/strong><\/p>\n\n\n\nGroups similar data points together without labeled data.<\/p>\n\n\n\n
from sklearn.cluster import KMeans\n\nkmeans = KMeans(n_clusters=3, random_state=42)\nclusters = kmeans.fit_predict(X)\nprint(\"Cluster centers:\", kmeans.cluster_centers_)<\/code><\/pre>\n\n\n\n5. Dimensionality Reduction (PCA)<\/strong><\/p>\n\n\n\nCompresses many correlated features into a smaller set that still captures most of the variance.<\/p>\n\n\n\n
from sklearn.decomposition import PCA\n\npca = PCA(n_components=2)\nreduced_data = pca.fit_transform(X)\nprint(\"Variance explained:\", pca.explained_variance_ratio_)<\/code><\/pre>\n\n\n\n6. Association Rule Mining (Apriori)<\/strong><\/p>\n\n\n\nFinds items that frequently occur together, the classic \u201ccustomers who bought X also bought Y\u201d model.<\/p>\n\n\n\n
from mlxtend.frequent_patterns import apriori, association_rules\n\nfrequent_itemsets = apriori(basket_df, min_support=0.05, use_colnames=True)\nrules = association_rules(frequent_itemsets, metric=\"lift\", min_threshold=1.0)\nprint(rules[[\"antecedents\", \"consequents\", \"support\", \"confidence\", \"lift\"]])<\/code><\/pre>\n\n\n\n7. Optimization Models (Linear Programming)<\/strong><\/p>\n\n\n\nFinds the best allocation of limited resources under a set of constraints.<\/p>\n\n\n\n
from scipy.optimize import linprog\n\nc = [-20, -30]\nA = [[1, 2], [3, 1]]\nb = [40, 60]\nresult = linprog(c, A_ub=A, b_ub=b, method=\"highs\")\nprint(\"Optimal values:\", result.x)<\/code><\/pre>\n\n\n\n8. Recommendation Systems<\/strong><\/p>\n\n\n\nSuggests items by finding users or items with similar behavior patterns.<\/p>\n\n\n\n
from sklearn.metrics.pairwise import cosine_similarity\n\nsimilarity_matrix = cosine_similarity(user_item_matrix)\nsimilar_users = similarity_matrix[0].argsort()[::-1][1:6]\nprint(\"Most similar users:\", similar_users)<\/code><\/pre>\n\n\n\n9. Reinforcement Learning Models<\/strong><\/p>\n\n\n\nLearns a policy by trial and error, updating its estimate of which action pays off best in each state.<\/p>\n\n\n\n
import numpy as np\n\nq_table = np.zeros((num_states, num_actions))\nlearning_rate = 0.1\ndiscount = 0.9\n\nfor episode in range(1000):\n state = env.reset()\n action = np.argmax(q_table[state])\n next_state, reward, done, _ = env.step(action)\n q_table[state, action] += learning_rate * (\n reward + discount * np.max(q_table[next_state]) - q_table[state, action]\n )<\/code><\/pre>\n\n\n\n10. Convolutional Neural Networks (CNN)<\/strong><\/p>\n\n\n\nBuilt for image data, learning spatial patterns like edges and shapes through convolutional layers.<\/p>\n\n\n\n
from tensorflow.keras import layers, models\n\ncnn_model = models.Sequential([\n layers.Conv2D(32, (3, 3), activation=\"relu\", input_shape=(28, 28, 1)),\n layers.MaxPooling2D((2, 2)),\n layers.Flatten(),\n layers.Dense(64, activation=\"relu\"),\n layers.Dense(10, activation=\"softmax\")\n])\n\ncnn_model.compile(optimizer=\"adam\", loss=\"sparse_categorical_crossentropy\", metrics=[\"accuracy\"])<\/code><\/pre>\n\n\n\n11. Recurrent Neural Networks (RNN) \/ LSTM<\/strong><\/p>\n\n\n\nBuilt for sequential data, carrying information forward from earlier steps to later ones.<\/p>\n\n\n\n
from tensorflow.keras import layers, models\n\nlstm_model = models.Sequential([\n layers.LSTM(64, input_shape=(timesteps, features)),\n layers.Dense(1)\n])\n\nlstm_model.compile(optimizer=\"adam\", loss=\"mse\")<\/code><\/pre>\n\n\n\n12. Transformers (BERT, GPT)<\/strong><\/p>\n\n\n\nUsed for language tasks, here a pretrained model classifying sentiment with no training required.<\/p>\n\n\n\n
from transformers import pipeline\n\nclassifier = pipeline(\"sentiment-analysis\")\nresult = classifier(\"This product completely changed how our team works.\")\nprint(result)<\/code><\/pre>\n\n\n
![\"How](\"https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/09\/How-Data-Science-Models-Work.jpg\" "\\"\\"")
<\/figure>\n\n\n\n