🚀 Create own ML Model and Sentiment analysis in iOS using Swift | by Pratiksha Mohadare | May, 2025

As cellular apps scale, backend server prices can shortly turn out to be a bottleneck. Luckily, iOS gadgets are getting smarter — and with the facility of on-device processing, we will shift important workloads to the consumer’s gadget, slicing infrastructure payments whereas enhancing efficiency and privateness. 💡

On this weblog, we’ll discover how you can optimize server prices by processing knowledge domestically on iPhones and iPads.

Right here’s why native knowledge processing is smart:

• Decrease server infrastructure prices 🏷️
• Offline performance 🌐🚫
• On the spot suggestions and low latency ⚡
• Enhanced privateness 🔒 (no must ship knowledge to servers)

Let’s construct a easy sentiment evaluation characteristic the place we classify user-generated textual content (like a evaluate or observe) utilizing a Core ML mannequin skilled with Create ML.

As a substitute of writing code or utilizing a CSV, you’ll simply set up just a few textual content recordsdata into folders and let Create ML do the magic. Right here’s how you can do it step-by-step.

You’ll create a folder that incorporates two subfolders:

SentimentData/
├── Pos/
│   ├── review1.txt     → "This app is wonderful!"
│   ├── review2.txt     → "I like utilizing this characteristic."
├── Neg/
│   ├── review1.txt     → "It is a dangerous expertise."
│   ├── review2.txt     → "I hate the brand new replace."

Every textual content file ought to include one brief assertion that displays both optimistic or destructive sentiment. You may create these recordsdata utilizing TextEdit or any easy textual content editor.

💡 The folder identify is the label. So Pos = Optimistic and Neg = Unfavourable.

1. Open the Create ML app on macOS.
2. Click on File > New Undertaking > Textual content Classification.
3. Identify your venture (e.g., “SentimentClassifier”).
4. Choose your SentimentData folder when requested for the coaching knowledge.
5. (Non-obligatory) Add a validation folder, or let it break up robotically.
6. Click on Prepare 🧠.
7. After coaching completes, click on Export and save your .mlmodel file.

• Drag and drop the exported .mlmodel file into your Xcode venture.
• Xcode auto-generates a category based mostly on the mannequin identify (e.g., SentimentClassifier).

import CoreML
/// Analyzes the sentiment of a given piece of textual content utilizing your customized mannequin.
func analyzeSentiment(textual content: String) -> String {
// Initialize the mannequin
guard let mannequin = attempt? SentimentClassifier(configuration: MLModelConfiguration()) else {
return "⚠️ Mannequin loading failed"
}
// Run prediction
guard let prediction = attempt? mannequin.prediction(textual content: textual content) else {
return "⚠️ Prediction failed"
}
return "✅ Sentiment: (prediction.label.capitalized)"
}

📌 Your mannequin will return labels like Pos, Neg, or no matter you named the folders.

import SwiftUI
struct SentimentView: View {
@State personal var inputText: String = ""
@State personal var sentimentResult: String = ""
var physique: some View {
VStack(spacing: 20) {
TextField("Kind your message...", textual content: $inputText)
.textFieldStyle(RoundedBorderTextFieldStyle())
.padding()
Button("Analyze Sentiment") {
sentimentResult = analyzeSentiment(textual content: inputText)
}
.padding()
.background(Shade.blue)
.foregroundColor(.white)
.cornerRadius(8)
Textual content(sentimentResult)
.font(.headline)
.padding()
}
.padding()
}
}

• Each consumer message despatched to the server
• Server runs ML inference
• Server sends again outcome ➡️ Excessive latency + value

• Message stays on gadget
• ML inference runs domestically
• Consequence proven immediately 🔥

Native knowledge processing is a sport changer. Not solely does it lower down on cloud bills, nevertheless it additionally empowers apps to be quicker and extra respectful of consumer privateness.

✅ Scale back prices
✅ Delight customers
✅ Adjust to privateness legal guidelines (like GDPR)

We’ll take a look at real-world case research the place on-device intelligence saved prices in large-scale apps. Keep tuned! 📊

The finished code for this venture together with .mlModel accessible right here
https://github.com/PratikshaMohadare/OnDeviceSentimentAnalysis-iOS-Swift